US5712861A  Error correcting method and decoder with improved reliability  Google Patents
Error correcting method and decoder with improved reliability Download PDFInfo
 Publication number
 US5712861A US5712861A US08501322 US50132295A US5712861A US 5712861 A US5712861 A US 5712861A US 08501322 US08501322 US 08501322 US 50132295 A US50132295 A US 50132295A US 5712861 A US5712861 A US 5712861A
 Authority
 US
 Grant status
 Grant
 Patent type
 Prior art keywords
 symbols
 error
 number
 errors
 rows
 Prior art date
 Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
 Expired  Fee Related
Links
Images
Classifications

 H—ELECTRICITY
 H03—BASIC ELECTRONIC CIRCUITRY
 H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
 H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
 H03M13/03—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words
 H03M13/05—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits
 H03M13/13—Linear codes
 H03M13/15—Cyclic codes, i.e. cyclic shifts of codewords produce other codewords, e.g. codes defined by a generator polynomial, BoseChaudhuriHocquenghem [BCH] codes
 H03M13/151—Cyclic codes, i.e. cyclic shifts of codewords produce other codewords, e.g. codes defined by a generator polynomial, BoseChaudhuriHocquenghem [BCH] codes using error location or error correction polynomials
Abstract
Description
This invention relates to a method and decoder for correcting errors in encoded digital information that has been transmitted over a communication channel or reproduced from a recording medium.
When such information is encoded, it is customary to add redundant check information that enables errors to be detected and corrected when the information is received or reproduced. Algebraic codes such as the ReedSolomon codes are often used. In these codes, the information to be transmitted or recorded is divided into symbols, each including of a certain number of bits, then into codewords, each including of a certain number of symbols. Additional check symbols are generated from each codeword by mathematical operations in which the information symbols are treated as the coefficients of a polynomial. The check symbols are appended to the codeword.
In the decoding process, the redundancy created by the check symbols enables errors in a codeword to be corrected, provided the number of symbols in error does not exceed half the number of check symbols. If more errors are present, the errors cannot be corrected, but it is usually possible to detect their presence, and flag the codeword as having uncorrectable errors.
Errorcorrecting capability can be improved by use of a twodimensional product encoding scheme such as the crossinterleave ReedSolomon code (CIRC) employed for recording digital audio slgnals on compact discs. and slmilar product codes often employed in video tape recording. The information to be encoded is divided into rectangular blocks of symbols. Column check symbols, referred to as C_{2} check symbols, are appended to vertical columns of information symbols to form C_{2} codewords; then row check symbols, referred to as C_{1} check symbols, are appended to horizontal rows of information symbols or C_{2} check symbols to form C_{1} codewords.
In the decoding process, first the C_{1} codewords are decoded. errors are corrected, and C_{1} codewords with uncorrectable errors are flagged. Next the C_{2} codewords are decoded, using the flags to identify likely error positions. Knowledge of the error positions enables a larger number of errors to be corrected: the number of correctable errors of known position per C_{2} codeword is equal to the number of C_{2} check symbols, instead of only half this number.
The total errorcorrecting capabllity of the C_{2} decoding process is given by the following inequality, in which e is the number of errors of unknown position, s is the number of errors of known position, and d_{2} is the minimum distance between C_{2} codewords (the minimum number of symbol positions in which any two correct C_{2} codewords differ; this number is one more than the number of C_{2} check symbols). If inequality (1) is satisfied, the errors can be corrected; if inequality (1) is not satisfied, the errors are uncorrectable.
2e+s<d.sub.2 ( 1)
If uncorrectable errors still remain after C_{2} decoding, for audio or video information, steps can be taken to conceal the errors, e.g. by substituting information obtained by interpolation from other codewords. Such error concealment can render the error substantially undetectable to a human listener or viewer. The embodiments of the present invention which will be described below are based on the above C_{1} C_{2} coding scheme, so it will be useful to explain this scheme in more detail. The mathematical operations by which check symbols are generated make use of a type of extended arithmetic in which each symbol is treated as an element in a Galois field. In many applications the Galols field is GF(2^{8}), in which each symbol includes of eight bits, but the principles are the same in any Galois field.
As an example, Table 1 lists the elements of GF(2^{4}), showing their representation both as fourbit symbols and as powers of a primitive element α. The four bits of each symbol represent the four coefficients a_{3}, a_{2}, a_{1}, and a_{0} of a polynomial of the form:
a.sub.3 z.sup.3 +a.sub.2 z.sup.2 +a.sub.1 z+a.sub.0
Addition and subtraction are carried out modulo two, by exclusiveOR logic operations on the coefficients. Addition is therefore identical to subtraction. Multiplication and division are carried out modulo a prime polynomial P(z), such as the following prime polynomial over GF(2^{4}):
P(z)+z.sup.4 +z+1
Thus α^{4} plus or minus α^{5} is α^{8} (0011+0110=0101), α^{4} multipiled by α^{5} is α^{9} (0011×0110=1010), α^{4} divided by α^{5} is α^{1} or α^{14} (0011/0110=1001), and α^{15} is equal to the unity element α^{0} (0001).
Since addition and subtraction are equivalent in any Galois field GF(2^{m}) with 2^{m} elements, the addition sign (+) and subtraction sign () can be used interchangeably in arithmetic operations over GF(2^{m}). Equations appearing hereinafter should be read with this in mind.
TABLE 1______________________________________GF (2.sup.4)Element Symbol Element Symbol______________________________________0 0000 α.sup.7 10111 (= α.sup.0) 0001 α.sup.8 0101α (= α.sup.1) 0010 α.sup.9 1010α.sup.2 0100 α.sup.10 0111α.sup.3 1000 α.sup.11 1110α.sup.4 0011 α.sup.12 1111α.sup.5 0110 α.sup.13 1101α.sup.6 1100 α.sup.14 1001______________________________________
Galols arithmetic operations and other operations required In encoding and decoding can be carried out at high speed by the hardware shown in FIGS. 37 to 40. Referring to FIG. 37, the hardware configuration includes a program unit 40, an arithmetic processor 50, and a Galois processor 80. The program unit 40 provides instructions to terminals T1 and T2. The instructions are executed by the arithmetic processor 50 and Galois processor 80, which are interconnected via further terminals T3 to T7. Additional data are input to the Galois processor 80 at a terminal T8. The Galois processor 80 is controlled by an external sequence controller (not shown), to which it is coupled by a terminal T9.
Referring to FIG. 38, the program unit 40 includes a branch address generator 41, a program counter 42, a readonly memory (ROM) 43, and an instruction latch 44. The program counter 42 generates successive addresses in the ROM 43, which outputs instructions stored at these addresses to the instruction latch 44. The instruction latch 44 provides the instructions to terminal T1 and to the branch address generator 41. Some of the provided instructions may be branch instructions, which cause the branch address generator 41 to set a new address in the program counter 42.
Referring to FIG. 39, the arithmetic processor 50 includes a pair of updown counters 55 and 56 and a pair of latches 57 and 58, the outputs of which are added by a pair of adders 59 and 60 to produce addresses at terminals T3 and T6. These addresses specify locations in two randomaccess memory areas (RAM A and RAM B) in the Galois processor 80. The counters 55 and 56 each provide, for example, five bits of output, specifying values from zero to thirtyone.
The arithmetic processor 50 also has a pair of flag registers 61 and 62 which indicate (by flags F_{3} and F_{4}) the status of counters 55 and 56, an arithmeticlogic unit (ALU) 63, another pair of flag registers 51 and 52 that indicate the status of results from the ALU 63, and a register file 64 with at least four registers, addressed as AD_{1} to AD_{4}. Input to register AD_{1} is controlled by the contents of a latch 65. Inputs to all the registers in the registers file 64 are selected by a selector 66. Two additional registers 67 and 68 hold data to be processed by the ALU 63.
These elements are interconnected as indicated by arrows in the drawings. The symbol A represents arithmetic result data from the ALU 63, Ca represents the value in counter 55. Cb represents the value in counter 56. P represents preset data received from terminal T4, and R represents register data from the register file 64.
Referring to FIG. 40, the Galois processor 80 includes the two randomaccess memory areas RAM A 81 and RAM B 82 mentioned earlier, a set of flag registers 84, 85, 86, 87, and 88, and a Galois arithmetic processing unit 89. This unit 89 consists of a Galoislogic unit (GLU) 90 that performs Galois arithmetic operations on three inputs G1, G2, and G3. and a reciprocal lookup table 91, stored in a readonly memory (ROM), that supplies the reciprocals of the elements in the Galois field, thereby facilitating division operations. The three inputs to the GLU 90 are obtained from an X register 92, a Y register 93, and a Z register 94. The operations performed by the GLU 90 include finding quantities such as X+Z, X/Y, XY+Z, and X^{2} Y+Z. Results are supplied at a terminal G0. The Galois processor 80 also has selectors 95, 96, and 97 that select inputs to RAM A 81 and RAM B 82.
The symbols A and B in FIG. 40 denote data read from RAM A 81 and RAM B 82, respectively. The letter I denotes, for example, syndrome data or Chien search data (described later). P and R denote preset data and register data as in FIG. 39, and V denotes Galois arithmetic results from the GLU 90. A gate 98 controlled by terminal T9 determines whether the I input is supplied from terminal T8 or selector 97. A flag register 99 stores a flag F5 indicating whether the result output by the GLU 90 is zero.
Flags set in registers 84, 85, 86, 87, and 88 include a computationinprogress flag St, an external bus read request flag EXR, an external bus write request flag EXW, and two flags F_{1} and F_{2} that are input from and output to the sequence controller (not shown).
Information including of k symbols a_{0}, . . . a_{k1} can be treated as the coefficients of a polynomial A(z), where
A(z)=a.sub.k1 z.sup.k1 +a.sub.k2 z.sup.k2 + . . . +a.sub.1 z+a.sub.0
To generate c check symbols, z^{c} A(z) is divided by a generator polynomial G(z) of degree c. The check symbols are the coefficients of the remainder B(z). For a ReedSolomon code, the generator polynomial G(z) has the following form, its roots being successive powers of a selected element β of the Galois field.
G(z)=(z+β.sup.j)(z+β.sup.j+1) . . . (z+β.sup.j+c1)
Except when stated otherwise, it will be assumed hereinafter that the roots of G(z) are the first c powers of a primitive element α of the Galois field, including the zeroth power or unity element, so that G(z) has the following form:
G(z)=(z+α.sup.0)(z+α.sup.1) . . . (z+α.sup.c1)
The codeword W with appended check symbols b_{0}, . . . , b_{c1} represents the polynomial z^{c} A(z)+B(z), or
W(z)=a.sub.k1 z.sup.k1+c + . . . +a.sub.0 z.sup.c +b.sub.c1 z.sup.c1 + . . . +b.sub.0
By construction, W(z) is divisible by G(z). Furthermore, the minimum distance d between codewords is equal to c+1; any two dlfferent codewords will differ In at least this number (c+1) of symbol positions. Thus d=c+1.
To decode a codeword W(z) that has been received from a communication channel or reproduced from a recording medium, W(z) is divided by G(z). Details will be omitted, but this polynomial division operation can be carried out rapidly by the hardware in FIGS. 37 to 40. The remainder is referred to as the syndrome polynomial S(z). If all coefficients of S(z) are zero, the codeword W(z) is regarded as correct. If all coefficients of S(z) are not zero, further decoding is performed to correct errors.
The first step in the errorcorrecting process is to apply the Euclidean algorithm for finding the greatest common divisor of two polynomials. The two polynomials are S(z) and z^{d1} (or z^{c}).
The execution of the Euclidean algorithm will be described with reference to FIGS. 37 to 43, assuming that there are sixteen check symbols, so that d=17, c=16, the generator polynomial G(z) is of degree sixteen, and the syndrome polynomial S(z) is of degree fifteen or less. No error location information will be used, so the maximum number of errors t that can be corrected is half the number of check symbols, or half the degree of G(z); in this cake the maximum number of correctable errors is eight.
In the general case, if d is the miningurn distance between codewords. and so d1 is the number of check symbols per codeword, then t is equal to the greatest integer not exceeding (d1)/2. Thls can be expressed as follows, using brackets to indicate the greatestinteger function.
t= (d1)/2!
The first step is to load the syndrome data. Counter 55. In the arithmetic processor 50 is initialized to fifteen. a value of sixteen is set in the register file 64 at addresss AD_{2} as the degree of the polynomial z^{d1}, and a base address is set in latch 56. Adder 59 generates the sum of the contents of counter 55 and the base address in latch 56, which is an address in RAM A 81. Latch 65 is cleared to zero, which enables data to be stored at address AD_{1} in the register file 64.
The syndrome coefficient data, which have been computed and stored in a memory area (not shown) are next read in at terminal T8, starting wlth the coefficient of z^{15} and routed via selector 95 to RAM A 81, where they are stored at addresses indicated by adder 59, As each coefficient symbol is stored, the contents (Ca) of counter 56 are loaded via selector 66 into the register file 64 at address AD_{1} ; then counter 56 is deeremented by one so that adder 59 indicates the next address.
When each coefficient symbol is loaded into RAM A 81, it is also sent to the GLU 89 and added to a preset value of zero. If the result of this addition is zero, indicating that the coefficient was the zero symbol, flag F_{5} in register 99 is cleared to zero. If the result is not zero, indicating a nonzero coefficient, flag F_{5} is set to one, and this flag value is latched in latch 65, disabling the storing of further data at address AD_{1} in the register file 64. At the end of the operation, address AD_{1} continues to hold the highest exponent of z with a nonzero coefficient in the syndrome S(z); this is the degree of the syndrome polynomial S(z).
Referring to the flowchart in FIG. 41, RAM A 81 and RAM B 82 are now initialized to start the Euclidean algorithm. Coefficient data for the polynomial z^{16} are stored as a polynomial M_{1} (z) in RAM B 82. The syndrome coefficients S(z) already stored in RAM A 81 form a second polynomial M_{2} (z). Two more polynomials U_{1} (z) and U_{2} (z) are initialized to zero, and a further polynomial U'_{1},(z) to the unity polynomial (1). U'_{1} (z) and U_{2} (z) are stored in RAM A 81 and RAM B 82, respectively. After tilts initialization, at point h_{0} in the flowchart, RAM A 81, RAM B 82, and the register file 64 contain the data shown at the left in FIG. 42.
Next M_{1} (z) is divided by M_{2} (z), using the GLU 90 in the Galois processor 80, to obtain a quotient polynomial Q(z) and remainder R(z). In the first iteration, M_{1} (z) is z^{16} and M_{2} (z) is S(z), so R(z) and Q(z) are related as follows:
R(z)=z.sup.16 Q(z)S(z) (2)
The coefficients of the remainder R(z) are stored in RAM B 82, replacing the M_{1} (z) data. Q(z), U'_{1} (z), and U_{2} (z) are used to calculate a new value for U_{1} (z), as follows:
U.sub.1 (z)=Q(z)U'.sub.1 (z)+U.sub.2 (z) (3)
Polynomial U_{2} (z) is then discarded, and polynomial U'_{1} (z) is reassigned as U_{2} (z). At this point j in the flowchart, RAM A 81, RAM B 82, and the register file 64 have the contents shown in the column headed j_{1} in FIG. 42. The degrees of polynomials M_{2} (z), R(z), U_{1} (z), and Q(z) are stored in the register file 64 at addresses AD_{1}, AD_{2}, AD_{3}, and AD_{4}, respectively.
The degree of the remainder R(z), stored in register AD_{2}, is now compared with the integer t, here equal to eight, designating the maximum number of correctable errors. If the degree of the remainder R(z) is equal to or greater than t, substitutions are made so that the old polynomial M_{2} (z) becomes the new M_{1} (z), the remainder R(z) becomes the new M_{2} (z), and U_{1} (z) becomes the new U'_{1} (z). These substitutions are made simply by exchanging the roles of RAM A 81 and RAM B 82, without actually transferring any data, as indicated in the column headed h_{1} in FIG. 42.
The algorithm now loops back to divide the new M_{1} (z) by the new M_{2} (z) and obtain a new quotient Q(z) and remainder R(z). Columns j_{2} and h_{2} lndleate the RAM and register contents at points j and h in the flowchart during this second iteration, assuming that the degree of the remainder R(z) is stlll equal to or greater than t. The symbols R_{1} (z) and R_{2} (z) indicate the remainder in the first and second iteration, respectively.
This process continues to loop until the degree of the remainder R(z) becomes equal to or less than t1. The roles of RAM A 81 and RAM B 82 are swapped at each iteration of the loop, so that the algorithm can be executed with a minimum of data transfer. FIG. 43 shows the RAM and register contents at points j and h in the flowchart for evennumbered iterations of the loop (columns j_{2m} and h_{2m}) and oddnumbered iterations of the loop (columns j_{2m+1} and h_{2m+1}).
When the degree of the remainder R(z) becomes t1 or less, the values of U_{1} (z) and R(z) are left stored as polynomials σ(z) and η(z), respectively, and the process ends. FIG. 43 shows the final state when the process ends in an even number of repetitions, so that σ(z) and η(z) are left stored in RAM A 81.
The degree of σ(z). stored at address AD_{3} in the register file, indicates the number of error symbols. If this degree exceeds t, the codeword is flagged as uncorrectable. An uncorrectable codeword is referred to as an erasure. If the degree does not exceed t, the roots of σ(z) are found by carrying out a Chien search; that is, by substituting successive powers of α in the Galois field for z, and testing whether the resulting value of σ(z) is zero. These roots designate the rotations off errors in the codeword. Details of the Chien search procedure will be given later (FIG. 21).
The polynomials σ(z) and η(z) can be normalized as follows, so that their leading coeficients are nnity:
σ(z)=U.sub.1 (z)=Kσ.sub.0 (z) (4)
η(z)=R(z)=Kη.sub.0 (z) (5)
If σ(z) has the maximum correctable degree t, it can be factored as follows: ##EQU1## The roots of σ(z) are α_{1}, α_{2}, . . . , α_{t}. Their values designate error symbols under a scheme by which, for example, α^{0} designates the coefficient of z^{0} in the codeword, α^{1} designates the coefficient of z^{1}, α^{2} designates the coefficient of z^{2}, and so on.
The difference between the codewords W(z) and W(z), referred to as the error pattern, gives the values that must be added to the symbols at the error locations to correct the errors. Like the codeword W(z), the error pattern is a collection of symbols; nonzero error pattern symbols correspond to symbols that are in error in W(z). Once the roots of σ(z) have been found. these nonzero error pattern symbols can be obtained by dividing η(z) by the formal derivative σ'(z), which includes of the odd terms of σ(z) with exponents of z reduced by one, setting z to each of the roots in turn. The error pattern symbol value e_{i} at the location corresponding to root α_{i} is given by:
e.sub.i =η(α.sub.i)/σ'(α.sub.i) (7)
Details will be omitted, but the Chien search and the computation of the error pattern values in equation (7) can be carried out rapidly on the hardware in FIGS. 37 to 40 by means of a suitable program stored in the ROM 43.
When error locations are already identified by erasure flags, the above procedure must be modified. A conventional algorithm which can be applied will be described below, using g(z) to represent the polynomial z^{d1}, which becomes an instance of a socalled Goppa polynomial.
When a received symbol r_{i} has been flagged as an erasure, its value is conventionally set to zero. If a_{i} was the transmitted value, then the value of the corresponding error pattern symbol becomes:
ε.sub.i =r.sub.i a.sub.i =a.sub.i ( 8)
If r_{i} is in error but has not been flagged as an erasure symbol, then the corresponding error pattern symbol has the following value:
e.sub.i =r.sub.i a.sub.i ≠0 (9)
The syndrome of the received codeword is calculated as follows, where n is the total number of symbols, including check symbols, in the codeword. ##EQU2##
Polynomials σ_{e} (z), η_{e} (z), φ_{e} (z), σ.sub.ε (z), η.sub.ε (z), and φ.sub.ε (z) are defined as follows, where E_{e} is the set error locations and E.sub.ε is the set of erasure locations. ##EQU3##
Polynomials S(z) and σ.sub.ε (z) can be calculated from the received codeword and erasure location information. The other polynomials are unknown, but they are related to S(z) and σ.sub.ε (z) by the following basic equation: ##EQU4##
If σ_{e} (z), η_{e} (z), and η.sub.ε (z) were known, the error locations E_{e} could be obtained by carrying out a Chien search to find the roots of σ_{e} (z), and the error pattern symbols e_{i} and ε_{i} could be found by means of formal derivatives, as follows: ##EQU5##
The decoding process is carried out by solving the basic equation (17). Three polynomials σ(z), η(z), and φ(z) are defined as follows:
σ(z)=σ.sub.e (z)σ.sub.ε (z) (20)
η(z)=η.sub.e (z)σ.sub.ε (z)+η.sub.ε (z)σ.sub.e (z) (21)
φ(z)=φ.sub.e (z)σ.sub.ε (z)+φ.sub.ε (z)σ.sub.e (z) (22)
The basic equation can then be rewritten as follows:
η(z)=σ(z)S(z)+φ(z)g(z) (23)
From the known polynomials S(z) and σ.sub.ε (z) it is possible to find two polynomials S.sub.ε (z) and φ.sub.ε (z) that satisfy the following conditions:
S.sub.ε (z)=σ.sub.ε (z)S(z)φ.sub.ε (z)g(z) (24)
deg S.sub.ε (z)<deg g=d1 (25)
Equation (23) can then be rewritten as follows:
η(z)=σ.sub.e (z)S.sub.ε (z)+ σ.sub.e (z)φ.sub.ε (z)+φ(z)!g(z) (27)
This modified basic equation (27) can be used for decoding by the Euclidean algorithm. The degrees (deg) of the polynomials appearing in equation (27) satisfy the conditions given by equations (28) to (30) below, and the greatestcommondivisor (gcd) condition given by equation (31). N_{e} is the number of errors, and N_{R} is the number of erasures. ##EQU6##
It can be shown that there is a unique set of polynomials {η(z), σ_{e} (z), σ_{e} (z)φ.sub.ε (z)+φ(z)} satisfying the basic equation (27). It can also be shown that the number of errors N_{e} is zero if and only if the degree of S.sub.ε (z) is less than the number of erasures N_{R}. It can furthermore be shown that if p.sub.ε (z) is the greatest common divisor of g(z) and S.sub.ε (z), the following inequality is satisfied:
deg p.sub.ε (z)<t+N.sub.R /2 (32)
An example using a ReedSolomon code with four check symbols over GF(2^{4}) will be given next. This generator polynomial of tills code is:
g(z)=z.sup.4 +z+1 (33)
The codeword length is twelve symbols, including eight information symbols and four check symbols. The transmitted codeword was the following:
v=(100 α0000 α.sup.12 0 α.sup.11 α.sup.9)
The received codeword, after erasure of a known incorrect symbol at the fourth position front the right, was:
r=(100 α.sup.9 000000 α.sup.11 α.sup.9)
Locations in these codewords are addressed by elements of GF(2^{4}) as follows, α^{11} indicating the leftmost location and α^{0} the rightmost location.
α.sup.11 α.sup.10 α.sup.9 α.sup.8 α.sup.7 α.sup.6 α.sup.5 α.sup.4 α.sup.3 α.sup.2 α.sup.1 α.sup.0
Incorrect received symbols occur at positions a^{8} and α^{3}, with α^{3} being a known erasure position and α^{8} an unknown error location. Front Table 1,
αα.sup.9 =α.sup.3
The error pattern value at location α^{8} is accordingly α^{3}. The erasure pattern value at location α^{3} is α^{12} . The following polynomials call be calculated.
σ.sub.ε (z)=z+α.sup.3 =z+α.sup.12( 34) ##EQU7##
Since the degree of S.sub.ε (z) is not less than the number of erasures, the number of errors of unknown location is not zero. The following relation, however, is satisfied:
(d1)/2+N.sub.R /2≦deg S.sub.ε (z) (37)
This indicates that the errors are correctable, so the Euclidean algorithm is used to find their locations. In the first iteration of this algorithm, z^{4} is divided by S.sub.ε (z) and the following relation is obtained:
z.sup.4 =(α.sup.8 z+1)S.sub.ε (z)+α.sup.3 z+α.sup.9 ( 38)
The degree of the remainder r_{1} =α^{3} z+α^{9} is one, so the iteration stops at this point, yielding the following results:
U.sub.1 (z)=q.sub.1 (z)=α.sup.8 (z+α.sup.8) (39)
σ.sub.e (z)=z+α.sup.8 =z+α.sup.7 ( 40)
η(z)=r.sub.1 =(α.sup.3 z+α.sup.9)α.sup.8 =α.sup.10 z+α (41)
σ(z)=σ.sub.e σ.sub.ε =(z+α.sup.7)(z+α.sup.12) (42)
σ'(z)=α.sup.7 +α.sup.12 =.sup.2 ( 43)
From equation (40), there is one error at location α^{8}. The error and erasure pattern values at locations α^{8} and α^{3} are found as follows: ##EQU8## The received codeword r is decoded by adding the error pattern e and erasure pattern ε as follows: ##EQU9##
The foregoing equations may make the erroranderasure decoding procedure appear complex, but it can be carried out with comparatively minor modifications to the flowchart in FIG. 41, by the hardware shown in FIGS. 37 to 40.
Although the coding and decoding schemes described above have been widely used, they are not without certain problems. One problem is that of miscorrection: a decoded C_{1} or C_{2} codeword may still contain errors, even though is not tagged as an erasure. This can happen when the number of errors originally present i the codeword exceeds the errorcorrecting capability of the code.
One conventional solution to this problem is to add further redundant information to the codewords. Typically a parity check symbol is added to each codeword before the check symbols are generated. By checking the parity symbol after the codeword has been decoded, it is possible to catch some miscorrection.
For a ReedSolomon code using a generator polynomial of the form G(z)=(z+α^{0})(z+α) . . . (z+α^{d2}), however, this parity scheme is far from satisfactory. A conventional parity check symbol is generated by adding all of the information symbols modulo two, by exclusiveOR logic, but the result obtained is equal to the remainder in polynomial division by (z+1), which has a root in common with G(z) (because α^{0} =1). Consequently, if a miscorrection is due entirely to errors in the information part of the codeword, and not in the check symbols themselves, the miscorrection will pass the parity check and will not be detected. More specifically, a parity check on the entire codeword, including check symbols, is calculated as follows: ##EQU10## where the r_{i} are the received symbols and the e_{i} are the error pattern symbols. Restricting the parity check to the k information symbols gives: ##EQU11## The correction to these symbols applied by the conventional decoding process will therefore cause the parity check to pass even if the corrections are incorrect.
A further problem of the conventional encoding scheme is that of overerasure. When a C_{1} codeword is flagged as an erasure, that does not mean that all the symbols in the codeword are incorrect. Typically, only a few of the symbols are incorrect. In the C_{2} decoding process it is not necessarily helpful to have correct symbols flagged as incorrect. This is particularly true under adverse conditions, when the symbol error rate is so high that the number of erasures exceeds the erasurecorrecting capability of the C_{2} code.
As a partial solution to this last problem, one of the present inventors has proposed a decoder that uses erasure information only when the number of erasures is within the correction capability of the C_{2} code. FIG. 44 shows the performance of this decoder. The symbol error rate prior to decoding is shown on the horizontal axis. The codeword error rate after decoding is shown on the vertical axis. The curve marked by crosses represents decoded codewords with uncorrected but detected errors. The curve marred with squares represents decoded codewords with wrongly corrected or undetected errors; that is, the number of miscorrected codewords.
The miscorrection curve trends generally downward as the symbol error rate decreases, but shows an upward jump in the circled area A. This corresponds to the symbol error rate at which it first becomes possible to make use of erasure information: the symbol error rate at which the number of erasures is typically just equal to the number of C_{2} check symbols. This jump in the curve indicates that erasure information is not being used in the most effective way to detect and correct errors.
Both of these problems are manifestations of the more general problem that conventional decoders do not provide an adequate check on the reliability of their results. Errors may slip through unrecognized, or correct results may be flagged as incorrect. Even when the results are correct and are so identified, it is not possible to place confidence in their correctness.
It is accordingly a general object of the present invention to improve the reliability of errorcorrecting codes.
Another general object is to assess the reliability of decoded results.
A more particular object is to detect miscorrections by polynomial codes.
A further object is to improve the errorcorrecting capability of product codes.
According to a first aspect of the present invention, a codeword contains information symbols, a parity check symbol, and additional check symbols. The parity check symbol is generated by treating the information symbols as coefficients of a polynomial over a Galois field, and evaluating this polynomial at a certain element of the Galois field. The additional check symbols are generated by treating the information symbols and parity check symbol as highorder coefficients of a polynomial over the Galois field, the loworder coefficients being zero, and dividing this polynomial by a generator polynomial not having the abovementioned certain element as a root.
According to a second aspect of the invention, the reliability of corrected codewords is assessed by counting the number of symbols found to be in error, or by counting the number of symbols having at least a certain number of bits in error. In decoding a rowcolumn array of symbols, in particular, first the rows are decoded, their reliability is assessed, and erasure flags are set to indicate unreliable rows; then columns are decoded, using the erasure flags to identify likely error positions. Use of the erasure flags can be made conditional on the number of erasure flags present; alternatively, column decoding can be initially attempted without using the erasure flags, then repeated as necessary with increasing numbers of erasure flags, the erasure flags being used in their reliability order.
According to a third aspect of the invention, when a rowcolumn array is decoded, rows and columns are assigned to a least reliable class, an intermediate class, and a most reliable class, reliability being assessed as in the second aspect. Rows belonging to the most reliable row class are corrected. Rows belonging to the intermediate row class are not corrected, but information about their errors is stored in a soft correction memory. Columns in the most reliable column class are corrected. Columns in the intermediate column class are corrected if their errors match the information in the soft correction memory at all positions. Other columns are flagged as having uncorrectable errors.
FIG. 1. illustrates the format of a product code with novel parity check symbols, used in a first embodiment: of the invention.
FIG. 2 is a block diagram of a paritychecking circuit for the code in FIG. 1.
FIG. 3 is a schematic diagram of the multiplier in FIG. 2.
FIG. 4 is a flowchart of the C_{2} decoding procedure in the first embodiment.
FIG. 5 is a graph illustrating the performance of the first embodiment.
FIG. 6 is a schematic diagram of another multiplier.
FIG. 7 is a schematic diagram of another multiplier.
FIG. 8 is a flowchart of a C_{2} decoding procedure in a second embodiment of the invention.
FIG. 9 illustrates the format of another product code with novel parity check symbols, used in a third embodiment.
FIG. 10 is a block diagram of an encoder used in the third embodiment.
FIG. 11 illustrates the encoding process in the third embodiment.
FIG. 12 illustrates a lookup table used in the third embodiment.
FIG. 13 shows an example of a lookup table over GF(2^{3}).
FIG. 14 shows the distribution of types of errors in symbols read from a video tape.
FIG. 15 shows the distribution of types of errors in simulated miscorrections of data received from a communication channel.
FIG. 16 is a block diagram of a novel decoder illustrating a fourth embodiment of the invention.
FIG. 17 is a block diagram of hardware for implementing the decoder In FIG. 16.
FIG. 18 is a block diagram of an arithmetic processor used in a fifth embodiment of the invention.
FIG. 19 is a flowchart of the C_{1} errorcorrecting process in the fifth embodiment.
FIG. 20 is a more detailed flowchart of the Euclidean algorithm as executed in FIG. 19.
FIG. 21 is a more detailed flowchart of a Chien search as executed in FIG. 19.
FIG. 22 illustrates memory and register contents during execution of the Euclidean algorithm in the fifth embodiment.
FIG. 23 illustrates further memory and register contents during execution of the Euclidean algorithm.
FIG. 24 is a block diagram of an arithmetic processor used in a sixth embodiment, of the invention.
FIG. 25 is a flowchart of the C_{2} decoding process the sixth embodiment.
FIG. 26 is a block diagram of a decoder used in seventh and eighth embodiments of the invention.
FIG. 27 illustrates a soft correction in the seventh embodiment.
FIG. 28 is a block diagram of a decoder used in ninth and tenth embodiments of the invention.
FIG. 29 is a block diagram of hardware for implementing the ninth embodiment.
FIG. 30 is a flowchart of the C_{1} decoding process in the ninth embodiment.
FIG. 31 illustrates C_{1} decoding results in the ninth embodiment.
FIG. 32 illustrates part of the C_{2} decoding process the ninth embodiment.
FIG. 33 is a flowchart of the C_{1} decoding process in the tenth embodiment.
FIG. 34 Illustrates C_{1} decoding results in the tenth embodiment.
FIG. 35 illustrates C_{2} preprocessing in the tenth embodiment.
FIG. 36 illustrating part of the C_{2} decoding process in the tenth embodiment.
FIG. 37 is a general block diagram of a conventional decoder.
FIG. 38 is a more detailed block diagram of the program unit in FIG. 37.
FIG. 39 is a more detailed block diagram of the arithmetic processor in FIG. 37.
FIG. 40 is a more detailed block diagram of the Galois processor in FIG. 37.
FIG. 41 is a flowchart of the Euclidean algorithm as executed by the decoder in FIG. 37.
FIG. 42 illustrates memory and register contents during execution of the Euclidean algorithm by the conventional decoder.
FIG. 43 illustrates further memory and register contents during execution of the Euclidean algorithm.
FIG. 44 is a graph illustrating the performance of a conventional decoder.
Several embodiments off the invention will now be described with reference to the attached illustrative drawings.
The first embodiment pertains to a product code over a Galois field GF(2^{8}) generated by the prime polynomial:
P(z)=z.sup.8 +z.sup.4 +z.sup.3 +z.sup.2 +1
FIG. 1 shows the code format. The information to be encoded is grouped into blocks of 76×138 eightbit symbols. Further check symbols are added to generate horizontal and vertical codewords. Each vertical codeword comprises one hundred thirtyeight information symbols and eleven C_{2} check symbols. Each horizontal codeword comprises seventysix information symbols or C_{2} check symbols, one parity check symbol (Cp), and eight C_{1} check symbols. The horizontal and vertical codewords will be referred to as C_{1} and C_{2} codewords, respectively.
The parity check symbol of a horizontal (C_{1}) codeword is obtained by the ReedSolomon encoding procedure, using a firstdegree generator polynomial z+α^{p}, where α is the element (00000010)of GF(2^{8}) and p is an integer less than zero or greater than seven. The seventysix information symbols are treated as coefficients of a polynomial A(z), and the parity check symbol is the remainder resulting from the division operation A(z)z/(z+α^{p}). Simple algebra shows that the parity check symbol has the value A(α^{p})α^{p}, so it can be obtained by evaluating A(z) at α^{p} to obtain an intermediate value, then multiplying this intermediate value by α.sup.α.
The C_{1} check symbols are obtained by treating the seventyseven symbols including the information symbols and parity check symbol as coefficients of a polynomial, multiplying by z^{8}, dividing by the following generator polynomial G_{1} (z), and taking the remainder.
G.sub.1 (z)=(z+α.sup.0)(z+α.sup.1) . . . (z+α.sup.7)
If the value of the parity check symbol is β, then the C_{1} check symbols are the coefficients of the remainder polynomial resulting from the following division operation:
A(z)z.sup.9 βz.sup.8 !/G.sub.1 (z)
The general requirement on p is that α^{p} must not be equal to any of the roots α^{0}, α^{1}, . . . , α^{7} of the generator polynomial G_{1} (z) of the C_{1} code. In the following description it will be assumed that α^{p} is a α^{1}, or α^{254}, representing the symbol (10001110).
The C_{2} check symbols are obtained by use of a generator polynomial G_{2} (z) of degree eleven on the first seventysix columns in FIG. 1.
In the decoding procedure, first C_{1} decoding is carried out as described previously, the eight C_{1} check symbols enabling correction of up to four symbols per codeword. C_{1} codewords found to have uncorrectable errors are flagged as erasures. C_{1} decoding can be carried out using the hardware shown in FIGS. 37 to 40. The information symbols and parity check symbol in a codeword output from C_{1} decoding will be referred to as a vector r, where
r=(r.sub.n1, r.sub.n2, . . . , r.sub.1, r.sub.0)
In the present embodiment, n equals seventyseven. The last symbol r_{0} is the parity check symbol.
Next a parity check is carried out on these symbols (r_{n1}, r_{n2}, . . . , r_{1}, r_{0}), by treating them as the coefficients of a polynomial, and determining whether this polynomial is evenly divisible by the parity generator polynomial (z+α^{1}). Equivalently, the parity check is carried out by evaluating this polynomial at α^{1} and testing whether the result is zero.
FIG. 2 shows a block diagram of a parity checker for carrying out this check, including a multiplier 1, a register 2, a modulotwo adder 3, a zero detection circuit 4, and a control circuit 5, interconnected as shown. The multiplier 1 multiplies the contents of the register 2 by α^{1}. The modulotwo adder 3 adds an input symbol r_{i} to the resulting product, and replaces the contents of the register 2 with the resulting sum. The zero detection circuit 4 tests whether the contents of the register 2 are zero. The control circuit 5 initializes the register 2, and has an internal counter (not shown) that controls the number of iterations of the above operations.
FIG. 3 shows a circuit that can be used as the multiplier 1. The eightbit contents "a" (a_{7}, a_{6}, . . . , a_{0}) of the register 2 are input at a terminal 6 and rotated one bit position to the right. Output bits b_{7}, b_{6}, b_{5}, b_{4}, and b_{0} have the values resulting from this rotation. Input bit a_{0} is added (modulo two) to bits a_{4}, a_{3}, and a_{2} by three exclusiveOR gates 7 to produce output bits b_{3}, b_{2}, and b_{1}. The output bits (b) are output at an output terminal 8 to the modulotwo adder 3 in FIG. 2.
To carry out the parity check, the control circuit 5 initializes the register 2 to zero. The counter in the control circuit 5 is initialized to n, the number of symbols to be checked (including the parity check symbol). Then the symbols are input one by one to the parity checker. Input of the first symbol r_{n1} places its value r_{n1} in the register 2. Input of the second symbol r_{n2} results in r_{n1} being multiplied by α^{1} and added to r_{n2}, placing the following sum, in the register 2.
r.sub.n1 α.sup.1 +r.sub.n2 (57)
Similarly, input of the next symbol r_{n3} places the following sum in the register 2:
(r.sub.n1 α.sup.1 +r.sub.n2)α.sup.1 +r.sub.n3(58)
When all n symbols, including the parity check symbol r_{0}, have been input, the control circuit 5 terminates the paritycheck operation. The final contents of the register 2, which are output as a parity syndrome S_{p}, are:
S.sub.p =(. . . (r.sub.n1 α.sup.1 +r.sub.n2)α.sup.1 +r.sub.n3)α.sup.1 + . . . +r.sub.1)α.sup.1 +rhd 0(59)
The parity syndrome S_{p} can also be written as follows, showing that it is equal to the polynomial with coefficients r_{i} evaluated at α^{1}. ##EQU12##
The zero detection circuit 4 tests the final S_{p} and sets or clears all erasure flag Fd. The erasure flag Fd is set if S_{p} is nonzero, indicating a parity error, and is cleared if S_{p} is zero, indicating a correctly decoded C_{1} codeword.
The parity check is followed by C_{2} decoding, which is again carried out by hardware like that shown in FIGS. 37 to 39. The C_{2} decoding process will be described with reference to the flowchart in FIG. 4.
The first step ST1 in the decoding of a C_{2} codeword is to determine whether erasure flags have been output by the C_{1} decoding process. The second step ST2 is to determine, from the flag values Fd output as above, whether the parity check failed for any C_{1} codewords after C_{1} error correction. If the answer in either of these steps is yes, erasure flags are set for the corresponding symbols in step ST3, but these symbols are not yet erased to zero.
The next step ST4 is to generate the C_{2} syndrome coefficients, as explained earlier. The next step ST5 is to determine whether the number of erasures, including both erasures flagged by C_{1} decoding and erasures flagged as a result of the parity check, is less than a certain number L. The value of L can be set equal to the minimum distance between C_{2} codewords, for example, so that L1 is the number of C_{2} check symbols (eleven in the present embodiment) and represents the maximum number of erasures that can be corrected.
If the number of erasures is less than L, then the erased symbols are set to zero, the syndrome polynomial is modified as explained earlier, and C_{2} decoding is carried out using the erasure information, in step ST6. If the number of erasures is equal to or greater than L, then the symbols flagged as erasures are not set to zero, and C_{2} decoding is carried out without using erasure information, in step ST7.
These C_{2} decoding processes may result in a determination of uncorrectable errors. This determination is made In the next steps ST8 and ST9, according to the degree of the syndrome polynomial S(z) or modified syndrome polynomial S.sub.ε (z). If uncorrectable errors are present, then the C_{2} codeword is flagged as an erasure in steps ST10 and ST11, so that error concealment measures can be taken.
In step ST12, if the entire block has been decoded, the decoding process ends. If not, the process returns to step ST1 to decode the next C_{2} codeword.
FIG. 5 illustrates the performance of the decoding procedure described above for various symbol error rates prior to C_{1} decoding. The horizontal and vertical axes and the curves marked by crosses and squares have the same meaning as in FIG. 44. The addition of the parity check can be seen to lead to a definite improvement over FIG. 44. The miscorrection curve shows no upward jump like that in area A in FIG. 44. For what is commonly taken to be a worstcase symbol error rate of 10^{2} the rate of miscorrected C_{2} codewords drops by about three orders of magnitude, from about 10^{20} in FIG. 44 to less than 10^{23} in FIG. 5, and the rate of detected but uncorrected errors C_{2} codewords (C_{2} erasures) shows a similar decline.
This improvement is due to the detection of miscorrected C_{1} codewords by the parity check. The parity check is particularly effective because the parity check is carried out using an element α^{1} that is not a root of the generator polynomial G_{1} (z) of the C_{1} code. A similar improvement could be obtained by using any other element α^{p} that is not a root of G_{1} (z).
With suitable modification of the multiplier circuit shown in FIG. 3, the parity checking circuit shown in FIG. 2 can be employed for other purposes, including C_{1} encoding and decoding. FIG. 6 shows a similar multiplier circuit structured to carry out multiplication by α, where as before, α is a root of z^{8} +z^{4} +z^{3} +z^{2} +1. Like the circuit in FIG. 3, the circuit in FIG. 6 has an input terminal 6, three exclusiveOR gates 7, and an output terminal 8. These elements are interconnected, however, so that the input bits a_{7} go a_{0} are rotated one big position to the left, with bit a_{7} being added to a_{3}, a_{2}. and a_{1} to generate output bits b_{4}, b_{3}, and b_{2}.
FIG. 7 shows a similar multiplier circuit structured to perform multiplication by α^{2}. This circuit requires four exclusiveOR gates 7, including two threeinput exclusiveOR gates. A threeinput exclusiveOR gate is configured by connecting a pair of twoinput exclusiveOR gates in series, so the circuit in FIG. 7 actually requires six exclusiveOR gates, twice as many as the circuits in FIGS. 3 and 6. The circuit in FIG. 7 is also slower, because of the propagation delays of the seriesconnected exclusiveOR gates.
Although various elements α^{p} can be selected for generating the parity check symbol, as FIGS. 6 and 7 illustrate, the selection makes a difference to the speed and complexity of the decoding (and encoding) circuits. Of all the elements in GF(2^{8}), α^{1} has the advantage of a multiplier circuit of minimum complexity and maximum speed. In the present embodiment the elements α^{0} and α^{1} cannot be used, because they are roots of the greater polynomial G_{1} (z) of the C_{1} code.
The invention can also be practiced with a C_{1} generator polynomial of the following form, where j is an integer greater than zero and d is the minimum distance between C_{1} codewords, so that d1 is the number of C_{1} check symbols:
(z+α.sup.j)(z+α.sup.j+1) . . . (z+α.sup.j+d2)
This makes it possible to use other elements, such as α^{0} or α^{1} to generate the parity check symbol. For example if the parity check symbol is generated by the conventional α^{0} and the C_{1} code by (z+α) (z+α^{2}) . . . (z+α^{d1}) then the parity checking operation and circuit become simpler. The operations required for C_{1} encoding and decoding, however, become more complex. Use of (z+α^{1}) as the generator polynomial for the parity check symbol and the conventional G_{1} (z) for the C_{1} code has the advantage of minimizing overall complexity.
Next a second embodiment of the invention will be described. This embodiment is similar to the first, except that the parameter L employed in C_{2} decoding is adjusted according to the number of miscorrections found by the parity check following C_{1} decoding. That is, the C_{2} decoding algorithm is switched according to the result of the parity check.
FIG. 8 shows the C_{2} decoding procedure of the second embodiment. In the first step ST21, all symbols in C_{1} 1 codewords identified as erasures by either C_{1} decoding or the parity check are flagged as such. In the next step ST22 the number Pe of parity errors is counted; that is, the number of C_{1} codewords which were not flagged as erasures in C_{1} decoding but failed the parity check is counted.
In the next step ST23 this parity error count Pe is compared with a threshold value La. If Pe is less than or equal to La, then in the next step ST24 the parameter L is set to a first value L_{0}. If Pe exceeds La, then L is set to a second value L_{1}, which is less than than L_{0}, in step ST25.
C_{2} decoding then proceeds as in the first embodiment. In step ST26 the syndrome coefficients are obtained. In step ST27 the total number of erasures flagged in step ST21 is compared with the value of L set as above. If the number of erasures is less than L, then the erasure information is used for erasure and error correction in step ST28. If the number of erasures is L or more, the erasure information is ignored and errorsonly correction is executed in step ST29. If uncorrectable errors are found in step ST30 or ST31 as a result of either C_{2} decoding process, then the C_{2} codeword is flagged as an erasure in step ST32 or ST33. Step ST34 decides whether the C_{2} codeword just decoded is the last in the block, and if it is not, returns to step ST21 to decode the next codeword.
When many C_{1} miscorrections are detected because they fail the parity check, this is generally symptomatic of a high symbol error rate in the originallyreceived data. By reducing the value of the parameter L in this situation and thereby reducing the maximum number of erasures that can be taken into consideration, the second embodiment reduces the errorcorrecting capability of the C_{2} code. As a result, the C_{2} decoding process will correct fewer errors and flag more C_{2} codewords as having uncorrectable errors. These codewords can then be dealt with by error concealment procedures such as interpolation.
If L_{0} is the minimum distance between C_{2} codewords, then when the number Pe of parity errors is within the threshold limit La, the full power of C_{1} C_{2} product decoding process is used, obtaining maximum error correction. When the number of parity errors exceeds La, indicating that the C_{1} decoding results are less reliable, less use is allowed of C_{1} erasure information, thereby avoiding C_{2} correction mistakes. The second embodiment accordingly provides improved reliability in the presence of a high symbol error rate, without sacrificing errorcorrecting capability when the symbol error rate is low.
FIG. 9 shows another possible location for the parity check symbols in the first two embodiments: at the head of the information symbols instead of between the information symbols and C_{1} check symbols. This is one of many possible variations on the position of the parity check symbols and the way in which they are generated. To mention another possible variation, the information symbols can be taken in one order for generating the parity check symbols and in another order for generating C_{1} check symbols: e.g. the parity check symbols can be generated by reading the information symbols in order from right to left in FIG. 9; then the C_{1} check symbols can be generated by reading the parity check and information symbols in order from left to right.
Next a third embodiment of the invention will be described. This embodiment uses the arrangement in FIG. 9, with the parity check symbols at the head of the information symbols.
In some data recording formats the first information symbol in each codeword is an identification (ID) symbol which may or may not be needed. If the ID symbol is present and is not needed, it can be advantageously used to hold the parity check information. This is one of the purposes of the third embodiment. Another purpose is to permit rapid encoding by generating parity check formation concurrently with C_{1} check information.
FIG. 10 shows an encoder for use in the third embodiment. Information symbols are received at an input terminal 9 and provided to a parity generator circuit 10 and a C_{1} information encoder 11. The C_{1} information encoder 11 generates a plurality of check symbols C_{1B} from the input information symbols, using a generator polynomial G_{1} (z) of the same type as in the first embodiment. The parity generating circuit 10 generates a parity check symbol Pr, using an element of the Galois field which is not a root of G_{1} (z). The parity check symbol is sent to a ROM lookup table 12 and expanded into a plurality of check symbols C_{1P}, which are combined by a modulotwo adder 13 with the C_{1B} check symbols output by the C_{1} information encoder 11. The parity check symbol Pr is also output at an output terminal 14. The sums of the C_{1P} and C_{1B} check symbols are output an output terminal 15 as row check symbols C_{1T}.
FIG. 11 illustrates the general encoding process. At the top of FIG. 11, check symbols C_{1B} and parity check symbols Pr are generated from a rowcolumn array in which each row includes of one ID symbol, having a fixed value of zero, and seventysix information symbols. The ID symbols are not used in generating the C_{1B} or Pr check symbols, or for any other purpose. The C_{1B} check symbols are temporarily appended to the right of the array, while the parity check symbols Pr are expanded into C_{1P} check symbols.
At the bottom of FIG. 11, the parity check symbols Pr have been substituted for the unused ID symbols, and now occupy the first, column of the array. The C_{1B} anti C_{1P} check symbols have been added by the modulotwo adder 13 and the resulting sums (C_{1T} check symbols) have been substituted for the C_{1B} check symbols at the right of the array.
The C_{1B} check symbols can be obtained from the k information symbols (0, a_{k2}, a_{k3}, . . . , a_{0}), including the initial zero ID symbol, by the usual C_{1} encoding procedure. The parity check symbol is obtained by feeding these k symbols (0, a_{k2}, a_{k3}, . . . , a_{0}) into, for example, the circuit of FIG. 2. In both cases, since the highestdegree coefficient is zero, it makes no difference whether this coefficient is included or not. The parity check symbol Pr is the remainder in the following division operation, and is equal to the dividend polynomial evaluated at α^{1}.
(a.sub.k2 z.sup.k2 +a.sub.k3 z.sup.k3 + . . . +a.sub.0)/(z+α.sup.1)
The parity check symbol Pr of the third embodiment differs from the parity check symbol Cp of the first embodiment in that evaluation of the dividend polynomial is not followed by multiplication by a α^{1}.
Referring to FIG. 12, the ROM lookup table 12 in FIG. 10 bas two hundred fiftysix addresses, corresponding to the possible values of the parity check symbol Pt. Eight C_{1P} check symbols (P_{7}, P_{6} . . . P_{0}) are stored for each address. The C_{1P} check symbols are obtained by accessing the ROM lookup table 12 at the address indicated by the Pr check symbol and reading out the C_{1P} data. These data are constructed so that when added to the C_{1B} data, they give the same result as if the C_{1} encoding procedure had been applied to information including the parity check symbol instead of the initial (zero) ID symbol.
The encoding and decoding procedures will now be explained through an example, using GF(2^{3}) instead of GF(2^{8}) for simplicity. Table 2 lists the elements of GF(2^{3}). The prime polynomial that generates this field is:
TABLE 2______________________________________P(z) = Z.sup.3 + Z + 1GF (2.sup.3)Element Symbol Element Symbol______________________________________0 000 α.sup.3 0111 (= α.sup.0) 001 α.sup.4 110α (= α.sup.1) 010 α.sup.5 111α.sup.2 100 α.sup.6 101______________________________________
The C_{1} codeword in this example has four information symbols (including the ID symbol) and three check symbols. The information symbols are (0, a_{2}, a_{1}, a_{0}), the initial zero being the unused ID symbol. The other symbols will have the following values in this example:
a.sub.2 =α.sup.2
a.sub.1 =α.sup.0 =1
a.sub.0 =α.sup.3
The parity symbol Pr is calculated as follows, by treating the information symbols as coefficients of a polynomial and evaluating this polynomial at α^{1}. ##EQU13##
The C_{1} check symbols will next be calculated in two ways, to show that both ways give the same result. First they will be calculated directly from the parity check symbol and information symbols, that is, from
(Pr, a.sub.2,a.sub.1, a.sub.0)=(α.sup.5, α.sup.2, α.sup.0, α.sup.3)
The C_{1} check symbols generated in this way will be denoted C_{10} =(P_{2}, P.sub., P_{0}). The resulting C_{1} codeword will be:
a=(α.sup.5, α.sup.2, α.sup.0, α.sup.3, p.sub.2, p.sub.1, p.sub.0)
From coding theory it is known that this codeword satisfies a matrix equation of the form:
aH.sup.T =0 (63)
In the present case this matrix equation is: ##EQU14## The check symbols C_{10} therefore satisfy the following simultaneous equations.
α.sup.5 +α.sup.2 +1+α.sup.3 +p.sub.2 +p.sub.1 +p.sub.0 =0(65)
α.sup.5 α.sup.6 +α.sup.2 α.sup.5 +α.sup.4 +α.sup.3 α.sup.3 +p.sub.2 α.sup.2 +p.sub.1 α+p.sub.0 =0 (66)
α.sup.5 α.sup.5 +α.sup.2 α.sup.3 +α+α.sup.3 α.sup.6 +p.sub.2 α.sup.4 +p.sub.1 α.sup.2 +p.sub.0 =0 (67)
These equations can be solved to obtain p_{2}, p_{1}, and p_{0}. Adding equations (65) and (66) gives:
α.sup.5 α.sup.6 +α.sup.2 α.sup.5 +α.sup.4 +α.sup.3 α.sup.3 +α.sup.5 +α.sup.2 +1+α.sup.3 +(1+α.sup.2)p.sub.2 +(1+α)p.sub.1 =α.sup.6 +α.sup.6 p.sub.2 +α.sup.3 p.sub.1 =0(68)
Adding equations (65) and (67) gives:
α.sup.5 α.sup.5 +α.sup.2 α.sup.3 +α+α.sup.3 α.sup.6 +α.sup.5 +α.sup.2 +1+α.sup.3 +(α.sup.4 +1)p.sub.2 +(α.sup.2 +1)p.sub.1 =α.sup.3 +α.sup.5 p.sub.2 +α.sup.6 p.sub.1 =0(69)
Multiplying equation (68) by α^{3} and adding equation (69) gives:
α.sup.9 +α.sup.3 +α.sup.9 p.sub.2 +α.sup.5 p.sub.2 =0
p.sub.2 =α.sup.2
Substituting this result into equation (69) gives:
α.sup.3 +α.sup.5 α.sup.2 +α.sup.6 p.sub.1 =0
p.sub.1 =α.sup.2
The results For P_{2} and P_{1} can be substituted into equation (65) to obtain P_{0}.
α.sup.5 +α.sup.2 +α.sup.0 +α.sup.3 +α.sup.2 +α.sup.2 +p.sub.0 =0
p.sub.0 =α.sup.0
The complete set of check symbols is thus:
C.sub.10 =(p.sub.2, p.sub.1, p.sub.0)=(α.sup.2, α.sup.2, α.sup.0)
Next, the same check symbols will be calculated by the method of the present embodiment. The first step is to calculate check symbols C_{1B} =(b_{2}, b_{1}, b_{0}) from the information symbols (0, a_{2}, a_{1}, a_{0}) without the parity check symbol. These check symbols can be found by solving the following matrix equation. ##EQU15## This yields the following simultaneous equations for the C_{1B} check symbols:
α.sup.2 +1+α.sup.3 +b.sub.2 +b.sub.1 +b.sub.0 =0(71)
α.sup.2 α.sup.5 +1α.sup.4 +α.sup.3 α.sup.3 +b.sub.2 α.sup.2 +b.sub.1 α+b.sub.0 =0 (72)
α.sup.2 α.sup.3 +1α+α.sup.3 α.sup.6 +b.sub.2 α.sup.4 +b.sub.1 α.sup.2 +b.sub.0 =0 (73)
Adding equations (71) and (72) gives:
α.sup.2 α.sup.5 +1α.sup.4 +α.sup.3 α.sup.3 +α.sup.2 +1+α.sup.3 +(α.sup.2 +1)b.sub.2 +(α+1)b.sub.1 =0
α.sup.2 +α.sup.6 b.sub.2 +α.sup.3 b.sub.1 =0(74)
Adding equations (71) and (73) gives:
α.sup.2 α.sup.3 +α+α.sup.3 α.sup.6 +α.sup.2 +α.sup.0 +α.sup.3 +(α.sup.4 +1)b.sub.2 +(α.sup.2 +1)b.sub.1 =0
α.sup.5 +α.sup.5 b.sub.2 +α.sup.6 b.sub.1 =0(75)
Multiplying equation (74) by α^{3} and adding equation (75) gives:
α.sup.3 α.sup.2 +α.sup.3 α.sup.6 b.sub.2 +α.sup.3 α.sup.3 b.sub.1 +α.sup.5 +α.sup.5 b.sub.2 +α.sup.6 b.sub.1 =0
(α.sup.2 +α.sup.5)b.sub.2 +0=0
b.sub.2 =0
Substituting this into equation (75) gives:
α.sup.5 +α.sup.5 x0+α.sup.6 b.sub.1 =0
b.sub.1 =α.sup.6
From equation (71), b_{0} is obtained as follows:
α.sup.2 +1+α.sup.3 +0+α.sup.6 +b.sub.0 =0
b.sub.0 =α.sup.2 +1+α.sup.3 +α.sup.6 =α.sup.3
The complete set of C_{1B} check symbols is thus:
C.sub.1B =(b.sub.2, b.sub.1, b.sub.0)=(0, α.sup.6, α.sup.3)
The C_{1P} check symbols are obtainable from the same matrix equation, using the vector (Pr, 0, 0, 0, q_{2}, q_{1}, q_{0}). ##EQU16## In the present example Pr is α^{5}, so:
α.sup.5 +q.sub.2 +q.sub.1 +q.sub.0 =0 (77)
α.sup.6 α.sup.5 +q.sub.2 α.sup.2 +q.sub.1 α+q.sub.0 =0 (78)
α.sup.5 α.sup.5 +q.sub.2 α.sup.4 +q.sub.1 α.sup.2 +q.sub.0 =0 (79)
Adding equations (77) and (79) gives:
(α.sup.5 +1)x α.sup.5 +(α.sup.4 +1)x q.sub.2 +(α.sup.2 +1)x q.sub.1 =0
α.sup.2 +α.sup.5 q.sub.2 +α.sup.6 q.sub.1 =0(80)
Adding equations (77) and (78) gives:
(α.sup.6 +1)x α.sup.5 +(α.sup.2 +1)x q.sub.2 +(α+1)x q.sub.1 =0
α.sup.0 +α.sup.6 q.sub.2 +α.sup.3 q.sub.1 =0(81)
Multiplying equation (81) by α^{3} and adding equation (80) gives:
α.sup.3 α.sup.0 +α.sup.3 α.sup.6 q.sub.2 +α.sup.3 α.sup.3 q.sub.1 +α.sup.2 +α.sup.5 q.sub.2 +α.sup.6 q.sub.1 =0
α.sup.3 +α.sup.2 +(α.sup.2 +α.sup.5)x q.sub.2 =0
q.sub.2 =α.sup.2
From equation (80),
α.sup.6 q.sub.1 +α.sup.2 +α.sup.5 α.sup.2 =0
q.sub.1 =α.sup.0
From equation (77).
α.sup.5 +α.sup.2 +α.sup.0 +q.sub.0 =0
q.sub.0 =α.sup.1
The complete set of C_{1P} check symbols is:
C.sub.1P =(q.sub.2, q.sub.1, q.sub.0)=(α.sup.2, α.sup.0, α.sup.1)
Adding the C_{1B} check symbols to the C_{1P} check symbols gives the same result as the C_{10} check symbols found earlier. ##EQU17##
To save time during the encoding process, the C_{1P} check symbols for each parity check symbol Pr are computed in advance and stored in the ROM lookup table 12. FIG. 13 shows the contents of the ROM lookup table 12 for the present example. This table was constructed by applying the procedure above to each possible value of Pr; that is, by substituting different values of Pr into the following matrix equation and solving for q_{2}, q_{1}, and q_{0}.
(Pr, 0, 0, 0, q.sub.2, q.sub.1, q.sub.0)x H.sup.T =0
This equation yields the following simultaneous equations:
Pr+q.sub.2 +q.sub.1 +q.sub.0 =0 (83)
Prα.sup.6 +q.sub.2 α.sup.2 +q.sub.1 α+q.sub.0 =0(84)
Prα.sup.5 +q.sub.2 α.sup.4 +q.sub.1 α.sup.2 +q.sub.0 =0(85)
Cramer's rule can be applied as follows to obtain q_{0} : ##EQU18## The symbol Δ denotes the following determinant: ##EQU19## The value of q_{0} is therefore: ##EQU20## Similarly, q_{1} and q_{2} can be found as follow: ##EQU21##
The table in FIG. 13 was thus generated from the formulas q_{2} =Prα^{4}, q_{1} =Prα^{2}, and q_{0} =Prα^{3}. A ROM lookup table 12 for GF(2^{8}) or any other Galois field can be constructed by similar formulas.
The parity and other check symbols in this embodiment can be calculated at high speed, as the information symbols are being input. As noted earlier, the parity check symbol Pr is calculated by using a circuit like that shown in FIG. 2 to work out the following equation (89), which evaluates a polynomial A(z) having the information symbols as coefficients when z=α^{1} ##EQU22##
The C_{1B} check symbols can be calculated by conventional encoding circuits, which are well known in the art. The final steps of looking up the C_{1P} check symbols in the ROM lookup table 12 and adding them to the C_{1B} check symbols take very little time. The encoding process of the fourth embodiment is therefore completed in substantially the same time as the conventional encoding process without a parity check symbol.
In the decoding process, conventional C_{1} decoding is applied to detect and correct errors in the C_{1} codewords. Next the parity check symbol Pr in each codeword is compared with a value calculated from the other information symbols In the codeword by the above equation (89). If the parity check symbol equals the calculated value, the parity check passes. Otherwise, the parity check fails and the codeword is flagged as having a parity error; that is, a C_{1} miscorrection is detected. C_{2} decoding then proceeds as in the first or second embodiment.
The third embodiment can be modified in various ways. For example, the lookup table for the C_{1P} check symbols can be stored in a randomaccess memory (RAM), or in a socalled flash memory, instead of in ROM. The generator polynomial of the C_{1} code need not be
G.sub.1 (z)=(z+α.sup.0) (z+α.sup.1) . . . (z+α.sup.d2)
It can have the more general form:
G.sub.1 (z)=(z+α.sup.j) (z+α.sup.j+1) . . . (z+α.sup.j+d2)
The parity check symbol need not be generated from α^{1} as in the preceding example, it can be generated from any nonzero element α^{p} of the Galois field that is not a root of the C_{1} generator polynomial G_{1} (z).
The third embodiment provides the same advantages as the first and second embodiments of reduced miscorrections and more reliable final results, with the additional advantage of an encoding method which is substantially as fast as the conventional encoding method.
In the preceding embodiments the reliability of C_{1} decoding results is improved by adding a parity check symbol to each C_{1} codeword. The reliability of C_{2} decoding results can be similarly improved by adding a parity check symbol to each C_{2} codeword. The C_{2} parity check symbol should be generated using an element of the Galois field that is not a root off the generator polynomial of the C_{2} code.
The reliability of C_{1} or C_{2} decoding results can be further improved by checking the number of bits corrected in each corrected symbol. First the reason for this will be explained, then further embodiments of the invention incorporating this principle will be described.
Consider, for example, information transmitted over a memoryless binary symmetric channel (n channel in which each bit has the same probability of being corrupted in transmission, with equal probabilities that a "0" will be changed to a "1" and that a "1" will be changed to a "0"). If the bit error rate is P_{b}, then the symbol error rate P_{s} is given by the following equation, in which _{8} C_{i} denotes the number off combinations of eight objects taken i at a time. ##EQU23## If the bit error rate P_{b} is 10^{3}, then the probability Ps_{1} of a onebit error in a symbol is: ##EQU24## Similarly, the probability of a twobit error is: ##EQU25## The probability of a threebit error is: ##EQU26## In comparison with the above probabilities, the probability of a fourbit error is so small as to be negligible.
Ps.sub.4 =.sub.8 C.sub.4 Pb.sup.4 (1Pb).sup.4 0
Similarly, the probability of a symbol having five or more bits in error is negligible.
Ps.sub.5 =Ps.sub.6 =Ps.sub.7 =Ps.sub.8 0
The same tendency is observed at all bit error rates P_{b} from 10^{2} to 10^{5}. In a memoryless binary symmetric channel, almost all symbol errors are onebit errors.
FIG. 14 shows the distribution of number of bits in error for symbols read from a digital video tape. These data were obtained by recording 220, 230 eightbit symbols with a 24/25 modulation scheme being. (In this modulation scheme, which is under consideration for use in a commercial digital video tape recorder, twentyfour data bits are recorded as twentyfive bits on the tape.) When the symbols were read from the tape and reproduced without error correction, fiftytwo symbols were in error, giving an average symbol error rate Ps of 2.86×10^{4}. Of these fiftytwo symbols, fiftyone had onebit errors, and one had a twobit error.
A somewhat similar situation occurs for signals recorded optically by eightfour modulation (EFM) on compact discs: here the majority of the errors are known to be twobit errors. For a wide variety of actual digital recording media and communication channels, the errors occurring in eightbit symbols are found to be predominantly onebit or twobit errors.
For errors due to miscorrection by a decoder, however, the situation is quite different. FIG. 15 5 shows results of a computer simulation of miscorrections by a ReedSolomon code over GF(2^{8}) with seventyseven information symbols and eight check symbols, assuming a received symbol error rate P_{s} of 10^{2}. One hundred thousand C_{1} codewords were decoded with correction of a maximum of four symbols per codeword, miscorrections were noted, and the number of corrected bits in each miscorrected symbol was counted. That is, the number of nonzero bits in the error pattern of each miscorrected symbol was counted. This number will be referred to as the bit weight of the error pattern of that symbol.
As FIG. 15 shows, the distribution of bit weights in miscorrected symbols is similar to a binomial distribution with a peak at the central value of four bits. Although not shown in FIG. 15, in symbols that were corrected correctly, the bit weight of the error pattern (number of bits corrected) was almost always one or two. Thus the error patterns of most miscorrected symbols had a bit weight of three or more, as indicated in FIG. 15, while the error patterns of correctly corrected symbols almost always had a bit weight of two or less.
Miscorrections can accordingly be identified with a fairly high degree of probability by looking for symbols in which more than two bits were corrected. The bit weight of the error pattern symbols can be used to provide a second check on the reliability of error correction, either in addition to or instead of the parity check described in the preceding embodiments.
A somewhat similar check can be obtained by considering the symbol weight of the error pattern, which is the number of symbols corrected (the number of nonzero error pattern symbols.) As a general rule, the greater the number of symbols corrected in a codeword is, the greater the chance of miscorrection of that codeword becomes. Thus a codeword in which four symbols have been corrected is more likely to have been miscorrected than a codeword in which only one symbol was corrected.
As a fourth embodiment of the invention, FIG. 16 illustrates a novel decoder adapted to assess the reliability of error correction in light of the bit weight and symbol weight of the error pattern. Symbols are received at an input terminal 21, stored in a memory 22, and corrected by an error detecting and correcting unit 23. The actual hardware structure of this error detecting and correcting unit 23 can be similar to the conventional structure shown in FIGS. 37 to 40, but conceptually, this structure can be considered to include of a syndrome calculation module 24, an error position calculation module 25, an error pattern calculation module 26, and an error correction module 27. The error detecting and correcting unit 23 is controlled by a decoding control circuit, 28.
In addition to the error detecting and correcting unit 23, the decoder in FIG. 16 provides a bitweight calculation circuit 29, a threshold register 30, a comparator 31, and a monitoring circuit 32, which are also controlled by the decoding control circuit 28. The monitoring circuit 32 outputs erasure flags FS at an output terminal 33, reliability information PM at an output terminal 34, and inhibiting signals at a terminal 35. These inhibiting signals disable the error correction module 27.
The bitweight calculation circuit 29 receives error patterns from the error pattern calculation module 26 at a terminal 36 and calculates the bit weight of each symbol in the error pattern, by counting the number of "1's." The comparator 31 compares the bit weight of each symbol in the error pattern with a threshold value L_{B} obtained from the threshold register 30, and notifies the monitoring circuit 32 of the result. For each codeword, the monitoring circuit 32 counts the number of symbols with bit weights greater than or equal to L_{B}. If this number exceeds a certain limit q, the monitoring circuit 32 sets an erasure flag FS for that symbol and sends an inhibiting signal to the error correction module 27 to prevent correction of that codeword.
When the error correction module 27 does not receive an inhibiting signal, it carries out the usual corrections on codewords stored in the memory 22. Corrected codewords, and codewords in which no errors were found, are output at a data output terminal 37.
Using error location information from the error position calculation module 25, for example, the decoding control circuit 28 counts the number of symbols corrected in each codeword and informs the monitoring circuit 32, which outputs a corresponding reliability value PM for the codeword at terminal 34. The reliability value may be based on the corrected symbol count alone, or may also take into account the number of error pattern symbols with bit weights of L_{B} or more, as reported by the comparator 31.
In the fourth embodiment (and other embodiments described herein) error correction can be intentionally restricted to reduce the probability of miscorrection. If the minimum distance between codewords is D, for example (i.e. the number of check symbols is D1), then for corrections of errors of unknown location, the maximum number of symbols corrected per codeword can be restricted to any number t_{a} not exceeding onehalf of D1; that is, to (D1)/2! or any smaller positive integer. The miscorrection probability is reduced if t_{a} is strictly less than (D1)/2!.
If error correction is restricted to t_{a} symbols, then the parameter q should have a value between one and t_{a} (1≦q<t_{a}). Appropriate selection of the parameters t_{a} and q can provide effective protection from miscorrections due to burst errors, as these miscorrections tend to produce correction of the maximum number t_{a} of symbols, with correction of a large number of bits in many of these symbols.
The erasure flags FS and reliability information PM output; by the monitoring circuit 32 in FIG. 16 can be used in various ways, several of which will be described in the following embodiments. Some of these embodiments use the erasure flags FS, or the erasure flags FS and reliability information PM, in decoding. Others use the bitweight or symbolweight information to make soft corrections; that is, to defer doubtful corrections until additional confirmation is received.
FIG. 17 is a general block diagram of the error detecting and correcting unit 23 used in the following embodiments. The program unit 40 and Galois processor 80 are similar to the conventional blocks with the same reference numerals in FIG. 37, except that the program and other data stored in the program unit 40 are adapted for carrying out functions of the decoding control circuit 28 in FIG. 16. For example, the values of the parameters L_{B} and q can be stored in the program unit 40.
The arithmetic processor 70 in FIG. 17 is similar to the arithmetic processor 50 in FIG. 37, but has additional facilities, which will be shown later, for making use of erasure and reliability information in C_{2} decoding. The arithmetic processor 70 outputs error pattern symbols at terminal 36, and receives inhibiting signals (INHI) at terminal 35. Terminals 35 and 36 are identical to the terminals with the same reference numerals in FIG. 16. Inhibiting signals INHI are also input to the program unit 40, where they cause a branch in program execution.
FIG. 18 shows the structure of the arithmetic processor 70 in the fifth embodiment. Elements identical to elements of the conventional arithmetic processor in FIG. 39 have the same reference numerals and perform the same functions. Only the new elements will be described below.
The arithmetic processor 70 has a shift register 71 that stores both received symbol data Din and erasure flags FS. Symbols flagged as erasures are indicated by shading in the drawing. The symbol data are stored in the memory 22 in FIG. 16. The erasure flags are stored in an erasure register 72. The erasure register 72 is an nbit register, where n is the number of symbols per C_{2} codeword. Erasures are indicated by setting the corresponding bit to "1" in the erasure register 72.
The erasure flags are counted by an erasure counter 73, which receives the contents of the erasure register 72 through an erasure control circuit 74. The count value in the erasure counter 73 is output to an address control circuit 75, which also receives an error count from the updown counter 55. From the error and erasure counts, the address control circuit 75 generates an address count, which is added to the value in latch 57 by adder 59 to generate a RAM address.
The erasure control circuit 74 is coupled to a Galois lookup table 76, wherein the erasure control circuit 74 converts the position of each erasure flag to an element of the Galois field. The addressing scheme is, for example, the scheme described earlier, in which successive positions are designated by α^{0}, α^{1}, α^{2}, . . . . The positions of up to d1 erasures and errors are stored in an erasure/error position register 77, which is organized as an 8×(d1)bit RAM. The parameter d1 is the number of C_{2} check symbols. Addresses in the erasure/error position register 77 are generated by the address control circuit 75 and adder 59 from the error and erasure counts provided by the counter 55 and erasure counter 73.
Next the operation of the fifth embodiment will be described.
In C_{1} decoding no erasure flags are received, and the arithmetic processor 70 In FIG. 18 operates like the arithmetic processor 50 In FIG. 39. The monitoring circuit 32 in FIG. 16 generates erasure flags for C_{1} codewords FS as described in the fourth embodiment. Additional erasure flags FS are generated if the error detecting and correcting unit 23 detects uncorrectable errors. Still further erasure flags FS may be generated by a parity check as described in the first embodiment.
In C_{2} decoding, all of these erasure flags are input to the shift register 71, together with C_{2} codewords read from the memory 22 in FIG. 16. The C_{2} decoding procedure will be described with reference to FIGS. 19 to 23.
Referring to FIG. 19, in the first step ST51 a C_{2} codeword is input and its syndrome polynomial S(z) is calculated. If all syndrome coefficients are zero, the codeword is presumed correct and output without correction.
In the next step ST52, the number of erasure flags (N_{R}) is counted by the erasure counter 73 and compared with the number of C_{2} check symbols (d1). If there are more erasure flags than C_{2} check symbols, the codeword is treated as uncorrectable and a branch is made to step ST59. described below. If there are not more erasure flags than C_{2} check symbols, the erasure control circuit 74 in FIG. 18 converts the erasure locations to elements α_{i} of the Galois field, using the lookup table 76, and stores these elements in the erasure/error position register 77.
In the next step ST53, the number of erasure flags compared with zero. If there are no erasure flags, the polynomial σ.sup.ε described in the background of the invention is set to unity (1) in step ST54. Otherwise, σ.sub.ε is set to the value given in equation (14); that is, to the product of terms of the form (zα_{i}), where the α_{i} (read from the erasure/error position register 77) designate the erasure positions.
In the next step ST55 a modified syndrome polynomial S.sub.ε (z) is calculated as the product of σ.sup.ε (z) and S(z), taken modulo z^{d1}.
In the next step ST56, the degree of the modified syndrome polynomial S_{e} (z) is compared with the number N_{R} of erasure flags. The degree of S.sub.ε (z) can be found in substantially the same way as the degree of the syndrome polynomial S(z) was found, as described in the background of the invention. If the number of erasure flags exceeds the degree of S.sub.ε (z), then all errors are erasures, and a branch is made to step ST61, described below.
If the number of erasure flags does not exceed the degree of S.sub.ε (z), then additional errors are present. In the next step ST57, the degree of S.sub.ε (z) is compared with the quantity (d1)/2+N_{R}. If the degree of S.sub.ε (z) is not less than (d1)/2+N_{R}, then the Euclidean algorithm and Chien search are executed in step ST58. This step will be described in more detail below. If the degree of S.sub.ε (z) is less than (d1)/2+N_{R}, the errors are uncorrectable and step ST59 is executed.
Step ST59 is an error concealment step in which the uncorrectable C_{2} codeword is replaced by, for example, data interpolated from preceding and following codewords. If N_{R} >d1 in step ST52, then an entire block of codewords is uucorrectable, and the entire block must be interpolated.
The Euclidean algorithm and Chien search as executed in step ST58 will be described with reference to FIGS. 20 23.
Referring to FIG. 20, the Euclidean algorithm begins with an initialization step SS1 in which a parameter i is set to unity, a polynomial M_{1} (z) is initialized to z^{d1}, and a polynomial M_{2} (z) is initialized to s.sub.ε (z). Additional polynomials U_{0} (z) and U_{1} (z) are initialized to one and zero, respectively.
In the next step SS2, M_{1} (z) is divided by M_{2} (z) to obtain a quotient Q(z) and remainder R(z), and a polynomial U_{i} (z) is set equal to Q(z)U_{i1} (z)+U_{i2} (z). In the first iteration, when i=1 a polynomial U_{1} (z) is set to Q(z)U_{0} (z)+U_{1} (z), which is equal to Q(z).
In the next step SS3, the degree of the remainder R(z) is compared with the quantity (d1)/2+N_{R}. If the degree of R(z) is greater than or equal to (d1)/2+N_{R}, then in step SS4 the parameter i is incremented by one, M_{1} (z) is replaced by M_{2} (z), and M_{2} (z) is replaced by R(z); then the procedure loops back to step SS2.
When the degree of the remainder R(z) is reduced to a value less than (d1)/2+N_{R}, a branch is made to FIG. 21 to execute the Chien search. In step SS5 a parameter j is initialized to unity, and a parameter k is initialized to 2^{m} 1, where 2^{m} is the number of elements in the Galois field (2^{8} in the present embodiment). In step SS6 the Galois field element α^{k} is tested to see if it is a root of the polynomial U_{i} (z). If it is, then in step SS7 α^{k} is stored in the erasure/error position register 77 as the location of the (j+N_{R})th error, and j is incremented by one. Next, in step SS8, parameter k is decremented by one. In step SS9 a test is made to see if α^{k} still designates a position in the information part of the C_{2} codeword. The lowercase letter k on the left of the inequality sign in step SS9 is the above parameter k. The uppercase K on the right of the inequality sign is the number of information symbols per C_{2} codeword. If k≧2^{m} K then α^{k} still designates a location in the information part of the codeword, and a return is made to step SS6 to find the next root of U_{i} (z).
FIG. 22 illustrates the contents of RAM A 81 and RAM B 82 in FIG. 40 and the register file 64 in FIG. 18 during the first two iterations of the Euclidean algorithm. Initially, M_{1} (z) and U_{1} (z) are stored in RAM B 82, and M_{2} (z) and U_{0} (z) in RAM A 81. After the first iteration, RAM A 81 holds U_{0} (z) and the new M_{1} (z), while RAM B 82 holds U_{1} (z) and the new M_{2} (z). After the second iteration, RAM A 81 holds M_{2} (z) and U_{2} (z), while RAM B 82 holds M_{1} (z) and U_{1} (z). That is, the same swapping system is used as in FIG. 42 to minimize actual data transfers.
FIG. 23 illustrates a general evennumbered iteration, a general oddnumbered iteration, and the final RAM and register contents. If the final iteration is even, the coefficients of the final polynomials R(z) and U_{i} (z) are left in RAM A 81. If the final iteration is odd, these final polynomial coefficients are left in RAM B 82.
Referring again to FIG. 19, after the execution of the Euclidean algorithm and Chien search, in step ST60 the polynomial σ(z) is set equal to U_{i} (z)σε(z), the polynomial η(z) is set equal to R(z), and a parameter L is set equal to N_{R} +j1. The parameter L is the total number of symbols to be corrected, including both erasures and other errors.
If the number of erasures exceeded the degree of the modified syndrome polynomial S.sub.ε (z) instep ST57, then in step ST61 polynomial σ(z) is set equal to σ.sup.ε (z), polynomial η(z) is set equal to S.sub.ε (z), and the parameter L is set, equal to N_{R}, indicating that all corrections are at erasure locations.
Following step ST60 or ST61, the error patterns of the symbols to be corrected are obtained in step ST62. If α_{i} designates the position of the ith incorrect symbol, its error pattern e_{i} is found as η(α_{i})/σ'(α_{i}), where as before σ'(α_{i}) is the value of the formal derivative of σ(z) α_{i}. This calculation is repeated for values of i from one to L to obtain an error pattern for each incorrect symbol.
Following step ST62, the C_{2} codeword is corrected by adding to each incorrect symbol the value of its error pattern symbol e_{i}. The parameter L (number of symbols corrected) can be output as a check on the reliability of the correction. The process then returns to step ST51 to correct the next C_{2} codeword.
The novel features of the fifth embodiment are the hardware features shown in FIG. 18 for handling erasure information during C_{2} decoding, and the use of bit weight information in FIG. 16 to generate erasure flags. By generating these additional erasure flags, the fifth embodiment strengthens the errorcorrecting capability of the C_{2} decoding process; that is, it pinpoints the locations of more errors. Increasing the number of erasure flags also has the effect of increasing the number of code blocks determined to be completely uncorrectable; that is, it produces a more conservative decoder that, when many errors are present, relies more on error concealment and less on error correction than does a conventional decoder.
The sixth embodiment uses the reliability information PM furnished by the monitoring circuit 32 in FIG. 16 to take a different approach to C_{2} decoding. FIG. 24 shows the hardware configuration of the arithmetic processor 70 in the sixth embodiment, using the same reference numerals to denote parts that are identical to the arithmetic processor 70 of the fifth embodiment In FIG. 18. Descriptions of these elements will be omitted.
The new elements in FIG. 24 are a reliability ordering circuit 78 and a reliability order register 79. The reliability ordering circuit 78 receives the reliability information PM from terminal 34. On the basis of this information it selects the least reliable d1 symbols and sorts them in order of reliability. The d1 reliability rankings thus generated are stored in the reliability order register 79 for output to the erasure control circuit 74. (The parameter d1 is the number of C_{2} check symbols.)
Equation (91) below shows an example of the way in which reliability values can be calculated for C_{1} codewords. ##EQU27##
The parameter D in tilts equation denotes the minimum distance between C_{1} codewords (the number of C_{1} check symbols is D1). If the jth C_{1} codeword is found to have errors at i locations, where i does not exceed (D1)/2, these errors are corrected and the jth codeword is assigned a weight W_{j} equal to i, the number of symbols corrected. If the codeword is found to have uncorrectable errors, its weight W_{j} is set equal to D/2. The reliability values of the C_{1} codewords are in reverse order to their weights, W_{j}, high weights W_{j} denoting low reliability and vice versa.
Reliability values can also be assigned on the basis of the bit weights employed in the fifth embodiment, e.g. by ranking C_{1} codewords according to the number of symbols having at least L_{B} symbols in error. Reliability rankings by bit weight can be used independently, or they can be used to refine the reliability rankings given by the symbol weights in equation (91). Another possible way to obtain reliability values is to count the total number of bits corrected in a C_{1} codeword. Reliability values may also be obtained from processes preceding C_{1} decoding. The invention is not limited to any particular method of obtaining reliability values.
The C_{1} decoding operation of the sixth embodiment identical to the C_{1} decoding operation of the fifth embodiment, so a description will be omitted. The operation of decoding a C_{2} codeword will be described with reference to FIG. 25.
In the first step ST41 in FIG. 25, the C_{2} syndrome polynomial S(z) is calculated and tested. If all its coefficients are zero, the C_{2} codeword is presumed to be errorfree.
In the next step ST42 the d1 least reliable symbols in the codeword are ranked by the reliability ordering circuit 78, and Galois field elements β^{1}, β_{2}, . . . , β_{d1} are stored in the reliability order register 79. β_{1} denotes the location of the least reliable symbol in the codeword, β_{2} the location of the nextleast reliable symbol, and so on. Symbols of equal reliability can be ranked in any order. If reliability is given in terms of the weights W_{j} in equation (90), which apply to C_{l} codewords, the reliability order will be the same for all C_{2} codewords in the same information block, so step ST42 need be performed only once per block.
In the next step ST43, the number of erasures N_{R} is initialized to zero, even if erasure flags were output for one or more C_{1} codewords in the C_{1} decoding step.
In the next step ST44, C_{2} decoding is attempted on the assumption that the least reliable N_{R} symbols in the codeword are erasures. The first time this step is carried out, the assumption is that there are no erasures (N_{R} is zero). The C_{2} decoding procedure in this step is identical to the procedure given in the fifth embodiment.
The next step ST45 is a decision step based on the result of the C_{2} decoding in step ST44. If the codeword was found to be correctable (if an error pattern was obtained), the correction is made and processing proceeds to the next codeword. If all uncorrectable error was detected, processing proceeds to step ST46.
In step ST46 an additional B symbols are treated as erasures. If {β^{1}, . . . , β_{i} } is the set of locations of symbols treated as erasures so far (i=NR), new locations B _{i+1}, . . . , β_{i+B} are now added to this set. The first time this step ST46 is carried out, the set {β_{1}, . . . , β^{i} } is empty (i=0), so the new set of erasures becomes {β_{1}, . . . , β_{B} }. The parameter B can be any positive integer less than d (1≦B ≦d1).
In the next step ST47, the value of N_{R} is updated by adding B.
In the next step N_{R} is compared with d1. If N_{R} does not exceed d1, a return is made to step ST44 to attempt C_{2} decoding again with more erasures than before. Each step ST44 is executed, the syndrome polynomial is modified to take account of the newly added erasure locations.
If N_{R} exceeds d1, an error concealment process is executed in step ST49.
One advantage of the sixth embodiment is that, in the first iteration of step ST44, no symbols are erased. As noted earlier, a C_{1} erasure flag does not imply that all symbols in the C_{1} codeword are in error; most of them are likely to be correct. By allowing these symbols to be treated as correct in the first iteration, the sixth embodiment improves the likelihood that the C_{2} decoding step will be able to correct other errors that were missed completely in the C_{1} decoding step.
A second advantage is that symbols of low reliability can be treated as erasures even if their C_{1} codewords were not flagged as erasures. The reliability information can thus be used to help locate miscorrections in the C_{1} decoding step, thereby improving the errorcorrecting capability of the C_{2} decoding step. If the number of C_{2} check symbols is eleven, for example, and if erasure flags were set on five C_{1} codewords, a conventional decoder could correct only three additional symbol errors per C_{2} codeword. By taking reliability information into account, however, the sixth embodiment is able to treat six more symbols as erasures, and correct tip to six additional symbol errors.
The next four embodiments employ the soft correction memory mentioned in the summary of the invention.
FIG. 26 shows the conceptual structure of a seventh embodiment of the invention, organized as a C_{1} decoder 100, soft correction memory 101, and C_{2} decoder 200. Both the C_{1} and C_{2} decoders incorporate the features shown in FIG. 16 for measuring the bit weights of error pattern symbols.
The C_{1} decoder 100 is allowed to correct a maximum of t_{1} errors per codeword, t_{1} being similar to the parameter t_{a} mentioned in the fourth embodiment, equal to at most half the number of C_{1} check symbols. Parameter values L_{B1} and q_{1} similar to the parameters L_{B} and q of the fourth embodiment (1≦q_{1} ≦t_{1}), are stored in the C_{1} decoder 100.
When decoded, C_{1} codewords fall into three reliability classes. Those with uncorrectable errors (errors in more than t_{1} symbols) belong to the least reliable class; these codewords are left uncorrected, and are flagged with erasure flags. C_{1} codewords with errors in t_{1} symbols or less are placed in the most reliable class if not more than q_{1} error pattern symbols have a bit weight of L_{B1} or more, and in the intermediate class if the bit weight of more than q_{1} error pattern symbols equals or exceeds L_{B1}. C_{1} codewords in the most reliable class are corrected, if they have any errors, and no erasure flags are output. C_{1} codewords in the intermediate class are not corrected, an erasure flag is output, and the error pattern and address of each incorrect symbol are stored in the soft correction memory 101. Storage of this information is referred to as a soft correction.
Referring to FIG. 27, the address of a symbol in a codeword is a pair of integers (x_{e}, y_{e}), in which x_{e} identifies the column and y_{e} identifies the row. The column address x_{e} indicates the C_{2} codeword and the row address y_{e} indicates the C_{1} codeword in which the symbol is disposed. For each soft correction, the address (x_{e}, y_{e}) of each symbol with a nonzero error pattern, and the value of the error pattern (e.g. e_{1} In FIG. 27) are stored in the soft correction memory 101. Storage of the C_{1} error pattern information in this form conserves memory space; depending on the C_{2} decoding algorithm, space for storing, at maximum, address and pattern information for only t_{1} ×t_{2} symbols suffices.
Referring again to FIG. 26, the C_{2} decoder 200 is allowed to correct up to t_{2} erasures, where t_{2} is, at maximum, the number of C_{2} check symbols. Parameter values L_{B2} and q_{2}, similar to the parameters L_{B} and q of the fourth embodiment (1≦q_{2} ≦t_{2}), are stored in the C_{2} decoder 200.
The erasure flags output by the C_{1} decoder 100 are used in C_{2} decoding. Like C_{1} decoding, C_{2} decoding sorts codewords into three reliability classes. C_{2} codewords found to have uncorrectable errors (e.g. erasures in more than t_{2} symbols) are placed in the least reliable class, their errors are not corrected, and an erasure flag is output at terminal 202, so that error concealment measures can be taken. C_{2} codewords with correctable errors are placed in the most reliable class if not more than q_{2} error pattern symbols have a bit weight of L_{B2} or more; these codewords are corrected (if they have any errors) and output at the data output terminal 201.
C_{2} codewords with correctable errors, but in which the error patterns of more than q_{2} symbols have a bit weight of L_{B2} or more, belong to the intermediate class. For these codewords, n comparison circuit 203 in the C_{2} decoder 200 compares each nonzero error pattern symbol with the information stored in the soft correction memory 101. If a matching error pattern value is found at a matching address, the correction is accepted as valid. If all corrections the C_{2} codeword are validated in this way, the codeword is corrected and output at the data output terminal 201. If any of the corrections are not validated, an erasure flag is output at terminal 202 to invoke error concealment processing.
The basic idea of the seventh embodiment is that if a correction fails the bitweight reliability test in either C_{1} or C_{2} decoding, the correction is made only if it is verified by occurrence of the same error pattern in both C_{1} and C_{2} decoding. Referring again to FIG. 27, suppose that a C_{2} codeword 204 is found to have correctable errors, but its error pattern fails the reliability test (more than q_{2} symbols with corrections to L_{B2} bits or more). A correction to a symbol with address (x_{e}, y_{e}) and error pattern e_{2} in this codeword is accepted as valid only if there was a soft correction at the same symbol address (x_{e}, y_{e}) in C_{1} decoding, and the C_{1} error pattern e_{1} at this address (x_{e}, y_{e}) was the same as the C_{2} error pattern (e_{1} =e_{2}). All of the symbols with nonzero error patterns e_{2} in the C_{2} codeword are checked in this way. If all corrections are validated, the codeword is corrected; otherwise it is flagged for error concealment processing.
The error pattern e_{2} that is compared with the error pattern e_{1} is the difference between the corrected symbol value and the original symbol value, before the symbol was erased to zero. With suitable alterations to the decoding algorithm, C_{2} error correction can be carried out using the original symbol values of erased symbols, instead of changing these values to zero. If C_{2} error decoding is carried out in the conventional way, however, with erased symbols cleared to zero, then the original symbol values must also be saved to enable meaningful comparison of their error patterns with the contents of the soft correction memory 101.
Compared with rejecting all C_{2} corrections that fail the reliability test with respect to L_{B2} and q_{2} the seventh embodiment provides the capability to correct more errors.
Next an eighth embodiment will be described. The eighth embodiment has the same structure as the seventh embodiment, with a soft correction memory 101 for storing soft correction information, and performs the same functions, with one additional function. After storing soft corrections for one block of received data, the C_{1} decoder 100 checks the soft correction memory 101 for the occurrence of soft corrections at a certain number q_{3} of consecutive symbol addresses. Consecutive means consecutive in the order in which the symbols were received from the communication channel or read from the recording medium. Due to interleaving, such consecutive addresses are not necessarily consecutive in the xy sense of FIG. 27.
The value of q_{3} can be any positive integer equal to or greater than two (for example q_{3} =3).
If soft corrections at q_{3} or more consecutive symbol addresses are found, they are accepted as valid, the assumption being that a burst error has occurred in the communication channel or on the recording medium. After all such occurrences of q_{3} or more consecutive soft corrections have been found, the number of symbols with bit weights of L_{B1} or more in the C_{1} codewords in which these soft corrections occurred is counted again, disregarding the burst errors. C_{1} codewords now pass the bitweight reliability test if they do not have corrections of L_{B1} bits or more in more than q_{1} symbols other than symbols involved in burst errors. Any C_{1} codewords now passing the bitweight reliability test are corrected, their erasure flags are removed, and their soft correction data are deleted from the soft correction memory 101.
C_{2} decoding is then carried out as in the seventh embodiment, using the remaining erasure flags and soft correction data. Compared with the seventh embodiment, the eighth embodiment permits still more errors to be corrected, by enabling C_{1} error correction to be employed to a greater extent on identifiable burst errors.
Next a ninth embodiment will be described. The ninth embodiment uses a soft correction scheme generally similar to that of the seventh and eighth embodiments, but based on symbol weights (number of symbols corrected per codeword) rather than bit weights. Also, the ninth embodiment does not use erasure information.
FIG. 28 shows the conceptual structure of the decoder in the ninth embodiment, using the same reference numerals as in FIGS. 16 and 26 to denote identical parts. In place of the circuits for monitoring bit weights in FIG. 16, the decoder in FIG. 28 has a soft correction memory 101 for storing soft correction patterns and addresses, as in FIG. 28. Error pattern symbols are obtained from the error pattern calculation module 26 via terminal 36, and their addresses from the error position calculation module 25 via a terminal 38. Error flags are output from the error correction module 27 at terminal 33. Inhibiting signals can be input to the error correction module 27 from terminal 35.
FIG. 29 is a general block diagram of the decoder hardware In the ninth embodiment, using the same reference numerals as In FIGS. 17 and 28 to denote equivalent or identical parts. The soft correction memory 101 is coupled to the arithmetic processor 70 through terminals 36 and 38 The program unit 40 and arithmetic processor 70 are interconnected through terminal 35 for passing inhibiting signals, which cause appropriate branches in the execution of the program in the program unit 40 and appropriate resetting of flags and register contents in the arithmetic processor 70.
FIG. 30 shows the C_{1} decoding procedure in the ninth embodiment. In step ST70 a C_{1} codeword is input, its syndrome coefficients are computed, and the usual C_{1} decoding operations are carried out to determine error positions and error patterns. The number of symbols in error will be denoted by the letter t, and the maximum errorcorrecting capability of the C_{1} code by t_{1}, where t_{1} is the greatest integer not exceeding half the number or C_{1} check symbols.
The next step ST72 is a decision step to determine whether the errors, if any, are correctable by the C_{1} code. If the codeword has uncorrectable errors (t>t_{1}), a branch is made to step ST76, described below.
in the next step ST72, if the codeword is correctable, the number of errors is compared with a parameter t_{a}. This parameter, described in connection with the fourth embodiment, is the maximum number of symbols that will actually be corrected. In the present embodiment t_{a} is an integer greater than zero but less than t_{1}. If the number of errors is equal to or less than t_{a}, a branch is made to step ST75, described below.
In the next step ST73, given that the number of errors is greater than t_{a} but not greater than t_{1} (t_{a} <t≦t_{1}), the addresses and error patterns of the symbols in error are stored in the soft correction memory 101, as in the seventh and eighth embodiments, thereby executing a soft correction.
Step ST73 is followed by step ST74, in which an inhibiting signal is output to prevent correction of the codeword. No erasure flag is output. C_{1} codewords processed in steps ST73 and ST74 belong to the intermediate reliability class.
If the number of errors is equal to or less than t_{a} (t≦Eta), step ST72 is followed by step ST75, in which the error pattern values (if any) are added to the codeword symbol values to correct the codeword. C_{1} codewords processed in step ST75 belong to the most reliable class.
If the codeword has uncorrectable errors (t>t_{1}), step ST71 is followed by step ST76, in which error correction is not performed, and no erasure flag is output. C_{1} codewords processed In step ST76 belong to the least reliable class.
FIG. 31 shows typical results of C_{1} decoding in the ninth embodiment: e_{xy} represents the error pattern of a symbol in error, the subscripts x and y being the column and row addresses of the symbol. The ith C_{1} codeword was found to have errors in symbols with column addresses p and q. The number of symbols in error did not exceed t_{a}, so the errors were corrected, as indicated by the dotted lines around their error patterns e_{ip} and e_{iq}. The jth C_{1} codeword was found to have uncorrectable errors (errors in more than t_{1} symbols), so although the errors were detected, no action was taken. The kth C_{1} codeword was found to have errors in a number of symbols exceeding t_{a} but not, exceeding t_{1}. These errors were not corrected, but information about them was stored in the soft correction memory 101 as a soft correction. The information stored includes the row address (k), column address (l, . . . , p, . . . , z), and error pattern (e_{kl}, . . . , e_{kp}, . . . , e_{kz}) for each symbol in error.
The C_{2} decoding process is similar to the C_{l} decoding process, employing parameters t_{2} and t_{b}, where t_{2} is the greatest integer not exceeding half the number of C_{2} check symbols and t_{b} is a positive integer less than t_{2}. Uncorrectable C_{2} codewords, with errors in more than t_{2} symbols, are placed in the least reliable class and flagged for error concealment. If errors are found in L_{b} symbols or less, the codeword is placed in the most reliable class and its error, if any, are corrected.
FIG. 32 illustrates the procedure for the intermediate class of C_{2} codewords, having errors in more than t_{b} but not more than t_{2} symbols. The pth C_{2} codeword is assumed to be such a codeword, with error patterns e_{jp}, . . . , e_{kp}, e_{mp}. Of these, only e_{kp} is stored in the soft correction memory 101.
The C_{2} decoding algorithm includes a step Z1 that checks each error in a C_{2} codeword in the intermediate class against the contents of the soft correction memory 101. If the same error pattern is found at the same address, the correction is accepted as valid. If all corrections in the C_{2} codeword are valid, the codeword is corrected (step Z2) and output (step Z3). If one or more corrections are not validated in the soft correction memory 101, the codeword is flagged as uncorrectable (step Z5) and error concealment is performed (step Z6).
In the present example, the pth C_{2} codeword is flagged as uncorrectable because error patterns e_{jp} and e_{mp} are not found in the soft correction memory 101. If, for example e_{mp} is a miscorrection, there is very little likelihood that a soft correction would also be stored by chance at the same address in the soft correction memory 101. Even if such a correction were stored, the chance that it would have the same error pattern is small. For GF(2^{8}), the probability of the error pattern matching by chance is only 1/256. In the present case, even if the error pattern and address of the e_{mp} were matched by chance, the codeword would still not be miscorrected, because e_{jp} is not present in the soft correction memory 101.
Since it does not use erasure information, the ninth embodiment does not offer high errorcorrecting capability, but it provides very strong protection from miscorrection.
Next a tenth embodiment will be described. The tenth embodiment is generally similar to the ninth, employs similar parameters t_{a}, t_{1}, t_{b}, and t_{2}, and has the structure shown in FIGS. 28 and 29, including the detailed structure of the arithmetic processor 70 shown in FIG. 18. The tenth embodiment differs from the ninth embodiment in that it uses erasure information.
FIG. 33 shows the C_{1} decoding procedure in the tenth embodiment. Steps ST70, ST71, ST72, and ST73 are the same as in the ninth embodiment, so descriptions will be omitted.
When a soft correction is performed in step ST73, an erasure flag is set in step ST77. The codeword is left uncorrected. The codeword belongs to the intermediate reliability class.
If the number of symbols in error (t) does not exceed parameter t_{a} in step ST72, the codeword is corrected in step ST78, and no erasure flag is set. The codeword belongs to the most reliable class.
If the codeword is found to be uncorrectable in step ST71 (t>t_{1}), an erasure flag is output in step ST79 and the codeword is left uncorrected. The codeword belongs to the least reliable class.
FIG. 34 shows typical results of C_{1} decoding in the tenth embodiment, including the erasure flags stored in the erasure register 72. As in FIG. 31, two errors have been corrected in the ith codeword, uncorrectable errors were detected in the jth codeword, and a soft correction was made on the kth codeword. Erasure flags have been output for the jth and kth codewords.
The C_{2} decoding process makes selective use or the erasure flags. Before C_{2} decoding begins, the number of erasure flags is compared with the parameter t_{2}, which now equal to, for example, the number of C_{2} check symbols (d_{2} 1). If the number of erasure flags exceeds t_{2}, an attempt is made to reduce the number of erasure flags by accepting some of the soft corrections.
Referring to FIG. 35, the total number S_{1} of erasure flags after C_{1} decoding is the sum of the number S_{c} of soft corrections and the number S_{e} of other erasure flags for C_{1} codewords with more than t_{1} errors). If S_{1} exceeds t_{2} and S_{e} is less than t_{2}, indicating that soft corrections have been made for at least two C_{1} codewords, then some (but not all) of these soft corrections are executed by adding the error pattern values stored in the soft correction memory 101 to the symbol values stored in memory 22. The C_{1} codewords for which soft corrections are executed at this point can be selected according to their reliability; for example, the C_{1} codewords with the fewest symbols in error can be corrected. Among C_{1} codewords of equal reliability, an arbitrary selection can be made.
The soft corrections that are executed at this stage are deleted from the soft correction memory 101, and their erasure flags are removed from the erasure register 72. If S_{a} is the number of C_{1} codewords that are corrected in this way, using the information in the soft correction memory 101, then only S_{b} soft corrections per C_{2} codeword are left stored in the soft correction memory 101 (where S_{a} +S_{b} =S_{c}), and only S_{2} erasure flags are left in the erasure register 72 (where S_{a} +S_{2} =S_{1}).
After this preprocessing step, C_{2} decoding is carried out as described in the ninth embodiment, except that the remaining erasure flags are used to identify error positions. The parameters t_{2} and t_{b} accordingly differ from the ninth embodiment: t_{2} (the maximum number of erasures that can be corrected) can be as large as the number of C_{2} check symbols (d_{2} 1), as noted above; t_{b} is a positive integer less than t_{2}.
If S_{2} is the number of erasures and E is the number additional errors discovered in decoding of a C_{2} codeword, and if the following inequality is satisfied, then the erasures and other errors are corrected unconditionally.
2E+S.sub.2 ≦t.sub.b
If this inequality is not satisfied, then the codeword is corrected only if E=0, meaning that only erasures are present, and only if there is no conflict between any of the corrections and the information stored in the soft correction memory. If any of the erasures are due to soft corrections, that is, the error patterns stored in the soft correction memory 101 for these erasures and the error patterns found in C_{2} decoding must be identical. If conflicting error patterns are found for any soft corrections, the C_{2} codeword is not corrected.
Referring to FIG. 36, suppose that the pth C_{2} codeword turns out to have errors at e_{jp}, e_{kp}, and e_{mp}, of which e_{jp} and e_{kp} have erasure flags, and e_{kp} has soft correction information stored in the soft correction memory 101. Suppose furthermore that the inequality given above is not satisfied. This codeword is not corrected, because e_{mp} does not have an erasure flag.
If e_{mp} were not present, then error patterns e_{jp} and e_{kp} would be checked against the contents of the soft correction memory 101 in step Z1 in the decoding process. For e_{jp} no conflict is possible because no soft correction is stored for address (p, j). A soft correction has been stored, however, for address (p, k). If the error pattern stored in the soft correction memory 101 for address (p, k) matched e_{kp}, the codeword would be corrected (step Z21) and output (step Z3). If the error pattern stored for address (p, k) did not match e_{kp}, the codeword would not be corrected, a flag would be set (step Z5) and error concealment measures would be taken (step Z6).
In FIG. 36 one soft correction e_{kp} was checked in pth codeword. As the Galois field GF(2^{8}) has two hundred fiftysix elements, there is a probability of1/256 that a miscorrection might generate a matching error pattern by chance. If two soft corrections are checked the probability that both checks might pass by chance is (1/256)^{2}, and if three soft corrections are checked the probability of chance agreement for all three error patterns is (1/256)^{3}. Checking just one soft correction will accordingly eliminate more than 99% of all C_{2} miscorrections, checking two soft corrections will eliminate more than 99.998%, and checking three soft corrections will eliminate more than 99.99999%. These figures give some idea of the power of the soft correction method.
Referring again to FIG. 35, in the preprocessing before C_{2} decoding, the number of soft corrections left, (S_{b}) should be at least one per C_{2} codeword, but S_{b} need not be a large number. As the above figures indicate, adequate protection against miscorrection is obtained if S_{b} is in the range from one to three. As long as this number of soft corrections is left, the number S_{a} of soft corrections that are executed in the preprocessing stage can be any number that reduces the number of erasure flags S_{2} to t_{2} or less (e.g. to d_{2} 1 or less).
The present invention is not limited to the preceding embodiments, but includes various further modifications to and combinations of these embodiments. In particular, the invention is not limited to twodimensional product codes; several of the embodiments are also useful for onedimensional codes (e.g. codes with only row cheer symbols), and all can be extended to product codes with more than two dimensions (e.g. codes with C_{1}, C_{2}, and C_{3} check symbols).
No restriction is placed on the number of symbols per codeword, which can be as large as the number of elements in the Galois field. No restriction is placed on the number of check symbols; the number of row check symbols need not be eight, and the number of column cheer symbols need not be eleven.
The Galois field GF(2^{8}) need not be generated by the prime polynomial P(z) given earlier. Other prime polynomials that can be used include the following two:
z.sup.8 +z.sup.5 +z.sup.3 +z.sup.2 +1
z.sup.8 +x.sup.7 +x.sup.2 +x.sup.1 +1
The invention can be practiced with Galois fields GF(2^{m}) where m is an integer other than eight, or more generally GF(q), where q is any power of any prime number. The conventional Galois field GF(2^{8}) is preferable, however, insofar as digital information is commonly expressed in eightbit bytes, and memory devices with an eightbit data width are readily available.
The invention is not limited to the particular hardware configurations described in the embodiments. For example, FIG. 16 was described as showing the conceptual structure of an embodiment having the actual hardware structure shown in FIGS. 17 and 18, but the invention can also be practiced in a decoder having separate hardware circuits for syndrome calculation, error position calculation, error pattern calculation, and error correction. Conversely, the control circuit, bit weight calculation circuit, comparator, and monitoring circuit in FIG. 16, which were described as separate hardware circuits, can be replaced by suitable program modules in the program unit in FIG. 17. Similarly, the C_{1} decoder and C_{2} decoder in FIG. 26 can be structured as separate software modules for C_{1} and C_{2} decoding, both executing on the same processor hardware.
Various modifications of an obvious nature can also be made to the flowcharts.
Those skilled in the art will recognized that further variations are possible without departing from the scope claimed below.
Further scope of applicability of the present invention will become apparent from the detailed description given hereinafter. However, it should be understood that the detailed description and specific examples, while indicating preferred embodiments of the invention, are given by way of illustration only, since various changes and modifications within the spirit and scope of the invention will become apparent to those skilled in the art from this detailed description.
Claims (60)
Priority Applications (8)
Application Number  Priority Date  Filing Date  Title 

JP18411094  19940712  
JP19845594  19940823  
JP6184110  19940922  
JP22830594  19940922  
JP6228305  19940922  
JP6198455  19940922  
JP7041941  19950301  
JP4194195A JP3328093B2 (en)  19940712  19950301  Error correction device 
Publications (1)
Publication Number  Publication Date 

US5712861A true US5712861A (en)  19980127 
Family
ID=27461125
Family Applications (1)
Application Number  Title  Priority Date  Filing Date 

US08501322 Expired  Fee Related US5712861A (en)  19940712  19950712  Error correcting method and decoder with improved reliability 
Country Status (2)
Country  Link 

US (1)  US5712861A (en) 
JP (1)  JP3328093B2 (en) 
Cited By (156)
Publication number  Priority date  Publication date  Assignee  Title 

US5898531A (en) *  19940901  19990427  Cannon Kabushiki Kaisha  Signal processing apparatus for digitally modulating input data 
US5901159A (en) *  19951225  19990504  Sony Corporation  Apparatus and method for decoding a coded digital signal 
US5983389A (en) *  19961008  19991109  Canon Kabushiki Kaisha  Error correction decoding apparatus 
US6026420A (en) *  19980120  20000215  3Com Corporation  Highspeed evaluation of polynomials 
US6029186A (en) *  19980120  20000222  3Com Corporation  High speed calculation of cyclical redundancy check sums 
US6047398A (en) *  19960703  20000404  Matsushita Electric Industrial Co., Ltd.  Reproducing method, reproducing apparatus and recording and reproducing apparatus using the same reproducing method, and recording medium having the same method recorded therein 
US6058500A (en) *  19980120  20000502  3Com Corporation  Highspeed syndrome calculation 
US6073250A (en) *  19971106  20000606  Luby; Michael G.  Loss resilient decoding technique 
US6108811A (en) *  19961018  20000822  Mitsubishi Denki Kabushiki Kaisha  Errorcorrecting decoder continuously adding flag signals to locations preceding a first location at which a difference between path metrics is lower than the threshold 
US6131178A (en) *  19970415  20001010  Mitsubishi Denki Kabushiki Kaisha  Error correcting decoding apparatus of extended ReedSolomon code, and error correcting apparatus of singly or doubly extended ReedSolomon codes 
US6138263A (en) *  19970408  20001024  Kabushiki Kaisha Toshiba  Error correcting method and apparatus for information data having error correcting product code block 
US6233710B1 (en) *  19970514  20010515  Texas Instruments Incorporated  ReedSolomon decoding device 
WO2002043250A2 (en) *  20001127  20020530  Nortel Networks France  Method for decoding a block of symbols and device therefor 
US6412090B1 (en) *  19980618  20020625  Globespanvirata, Inc.  Galois field computation system and method 
JP3328093B2 (en)  19940712  20020924  三菱電機株式会社  Error correction device 
US20030140301A1 (en) *  20020123  20030724  Litwin Louis Robert  Intradecoder component block messaging 
US20030172328A1 (en) *  20020306  20030911  Wyatt Stewart R.  Pausing a transfer of data 
US6625774B1 (en) *  19991029  20030923  Stmicroelectronics, Inc.  Redundancy system and method for locating errors in interleaved code words 
US6634007B1 (en)  19991108  20031014  Codevector Technology  Algebraic soft decoding of reedsolomon codes 
US20040003334A1 (en) *  20020630  20040101  Boyanapalli Vipin S.  Efficient method and apparatus for low latency forward error correction 
US20040030737A1 (en) *  20020812  20040212  International Business Machines Corporation  Algebraic decoder and method for correcting an arbitrary mixture of burst and random errors 
US20040059985A1 (en) *  20020925  20040325  Sharp Guy Bl  Method and apparatus for tracking address of memory errors 
US6721917B2 (en) *  20000616  20040413  Acer Laboratories, Inc.  Method and system for optical disk decoding 
US6721373B1 (en) *  20000329  20040413  Tioga Technologies Ltd.  Multitone receiver and a method for operating the same 
US20040078407A1 (en) *  20021017  20040422  Mats Naslund  Efficient arithmetic in finite fields of odd characteristic on binary hardware 
US20040078745A1 (en) *  20010218  20040422  Martin Kruger  Method and device for error correction of data blocks 
US20040193966A1 (en) *  20030313  20040930  Osamu Hirabayashi  Semiconductor device 
US20040194003A1 (en) *  20000609  20040930  Kwon HyungJoon  Method and apparatus for correcting C1/PI word errors using error locations detected by EFM/EFM+ decoding 
EP1468498A1 (en) *  20020123  20041020  Thomson Licensing S.A.  Chien search cell for an errorcorrecting decoder 
EP1470647A1 (en) *  20020123  20041027  Thomson Licensing S.A.  Dual chien search blocks in an errorcorrecting decoder 
US20040249840A1 (en) *  20030603  20041209  Sunplus Technology Co., Ltd.  Device and method for block code error correction 
US20050015706A1 (en) *  19971201  20050120  Macrovision Europe Limited  Codeword for use in digital optical media and a method of generating therefor 
US20050120285A1 (en) *  20031128  20050602  Mediatek Inc.  Encoding method and apparatus for cross interleaved cyclic codes 
US20050229085A1 (en) *  20020405  20051013  Koninklijke Philips Electronics N.V.  Method and apparatus for embedding an additional layer of error correction into an error correcting code 
WO2005101207A1 (en) *  20040414  20051027  Koninklijke Philips Electronics N.V.  Data handling device that corrects errors in a data memory 
US20050289444A1 (en) *  20040625  20051229  Klein Dean A  Low power costeffective ECC memory system and method 
US20060056259A1 (en) *  20040506  20060316  Klein Dean A  Memory controller method and system compensating for memory cell data losses 
WO2006068436A1 (en) *  20041222  20060629  Samsung Electronics Co., Ltd.  Apparatus and method for decoding reedsolomon symbols 
US20060150054A1 (en) *  20050103  20060706  JiaPing Chen  Decoding device for decoding product code and decoding method using the same 
US20060174182A1 (en) *  20050128  20060803  Neefs Henk G C  Trialanderror multibit error correction 
US20060206769A1 (en) *  20040624  20060914  Klein Dean A  Memory system and method having selective ECC during low power refresh 
US20060248430A1 (en) *  20050428  20061102  Sandbridge Technologies Inc.  Iterative concatenated convolutional ReedSolomon decoding method 
US7159164B1 (en) *  20000605  20070102  Qualcomm Incorporated  Method and apparatus for recovery of particular bits of a frame 
US20070150794A1 (en) *  20021017  20070628  Mats Naslund  Error correction using finite fields of odd characteristic on binary hardware 
US20070233984A1 (en) *  20060329  20071004  Stmicroelectronics S.R.L.  Method and device for detecting possible corruption of sector protection information of a nonvolatile memory stored in an on board volatile memory array at poweron 
US20080002503A1 (en) *  20040715  20080103  Klein Dean A  Method and system for controlling refresh to avoid memory cell data losses 
US20080016432A1 (en) *  20060712  20080117  Peter Lablans  Error Correction in MultiValued (p,k) Codes 
US20080092016A1 (en) *  20061011  20080417  Micron Technology, Inc.  Memory system and method using partial ECC to achieve low power refresh and fast access to data 
WO2008045292A2 (en) *  20061004  20080417  Marvell World Trade Ltd.  Error correction decoding methods and apparatus 
US20080109705A1 (en) *  20061018  20080508  Pawlowski J Thomas  Memory system and method using ECC with flag bit to identify modified data 
US20080109687A1 (en) *  20061025  20080508  Christopher Michael Abernathy  Method and apparatus for correcting data errors 
US20080115041A1 (en) *  20061114  20080515  Jihoon Park  Error correction method and apparatus for predetermined error patterns 
US20080168330A1 (en) *  20070108  20080710  Agere Systems Inc.  Systems and methods for prioritizing error correction data 
US20080320358A1 (en) *  20040727  20081225  Jurgen Pandel  Encoding and Decoding Method, and Encoding and Decoding Devices with a TwoStage Error Protection Process 
US20090083590A1 (en) *  20070926  20090326  John Johnson Wylie  System and method for determining the faulttolerance of an erasure code 
WO2009041153A1 (en) *  20070926  20090402  Kabushiki Kaisha Toshiba  Semiconductor memory device and its control method 
US20090172501A1 (en) *  20060303  20090702  Ternarylogic Llc  MultiState Symbol Error Correction in Matrix Based Codes 
US20090228769A1 (en) *  20050203  20090910  Vityaev Andrei E  Metaviterbi algorithm for use in communication systems 
US20090249164A1 (en) *  20060802  20091001  Jorg Hammer  Method for serial asynchronous transmission of data in an arragement for the monitoring, controlling, and regulating an operational control facility of building 
US20090282313A1 (en) *  20080507  20091112  Legend Silicon Corp.  Method and apparatus for hard decision bounded distance decoding of nordstrom robinson code 
US20090292975A1 (en) *  20050429  20091126  Agere Systems Inc.  Method and apparatus for iterative errorerasure decoding 
US20100083069A1 (en) *  20080618  20100401  John Johnson Wylie  Selecting Erasure Codes For A Fault Tolerant System 
US20110047441A1 (en) *  20080301  20110224  Kabushiki Kaisha Toshiba  Chien search device and chien search method 
US20110080211A1 (en) *  20081120  20110407  Shaohua Yang  Systems and Methods for Noise Reduced Data Detection 
US20110164669A1 (en) *  20100105  20110707  Lsi Corporation  Systems and Methods for Determining Noise Components in a Signal Set 
US8037379B1 (en) *  20060831  20111011  Pdf Solutions, Inc.  Prediction of impact on postrepair yield resulting from manufacturing process modification 
US8161351B2 (en)  20100330  20120417  Lsi Corporation  Systems and methods for efficient data storage 
US8208213B2 (en)  20100602  20120626  Lsi Corporation  Systems and methods for hybrid algorithm gain adaptation 
CN102656566A (en) *  20091217  20120905  国际商业机器公司  Data management in solid state storage systems 
US8295001B2 (en)  20100921  20121023  Lsi Corporation  Systems and methods for low latency noise cancellation 
US8359522B2 (en)  20070501  20130122  Texas A&M University System  Low density parity check decoder for regular LDPC codes 
US8381071B1 (en)  20100521  20130219  Lsi Corporation  Systems and methods for decoder sharing between data sets 
US8381074B1 (en)  20100521  20130219  Lsi Corporation  Systems and methods for utilizing a centralized queue based data processing circuit 
US8385014B2 (en)  20101011  20130226  Lsi Corporation  Systems and methods for identifying potential media failure 
US8413020B2 (en)  20090812  20130402  Lsi Corporation  Systems and methods for retimed virtual data processing 
US8418019B2 (en)  20100419  20130409  Lsi Corporation  Systems and methods for dynamic scaling in a data decoding system 
US20130110962A1 (en) *  20111101  20130502  Cleversafe, Inc.  Wirelessly Sending a Set of Encoded Data Slices 
US8443250B2 (en)  20101011  20130514  Lsi Corporation  Systems and methods for error correction using irregular low density parity check codes 
US8443249B2 (en)  20100426  20130514  Lsi Corporation  Systems and methods for low density parity check data encoding 
US8443251B1 (en)  20111215  20130514  Lsi Corporation  Systems and methods for out of order processing in a data retry 
US8443271B1 (en)  20111028  20130514  Lsi Corporation  Systems and methods for dual process data decoding 
US8446683B2 (en)  20110222  20130521  Lsi Corporation  Systems and methods for data precoding calibration 
US8468418B2 (en)  20080502  20130618  Lsi Corporation  Systems and methods for queue based data detection and decoding 
US8479086B2 (en)  20111003  20130702  Lsi Corporation  Systems and methods for efficient parameter modification 
US8499231B2 (en)  20110624  20130730  Lsi Corporation  Systems and methods for reduced format nonbinary decoding 
US8525707B1 (en)  20120517  20130903  Lsi Corporation  Systems and methods for dual binary and nonbinary decoding processing 
US8527858B2 (en)  20111028  20130903  Lsi Corporation  Systems and methods for selective decode algorithm modification 
US8527831B2 (en)  20100426  20130903  Lsi Corporation  Systems and methods for low density parity check data decoding 
US20130232396A1 (en) *  20120305  20130905  Micron Technology Inc.  Apparatuses and methods for encoding using error protection codes 
US8531320B2 (en)  20111114  20130910  Lsi Corporation  Systems and methods for memory efficient data decoding 
US8539328B2 (en)  20110819  20130917  Lsi Corporation  Systems and methods for noise injection driven parameter selection 
US8560930B2 (en)  20101011  20131015  Lsi Corporation  Systems and methods for multilevel quasicyclic low density parity check codes 
US8560929B2 (en)  20110624  20131015  Lsi Corporation  Systems and methods for nonbinary decoding 
US8566666B2 (en)  20110711  20131022  Lsi Corporation  Minsum based nonbinary LDPC decoder 
US8566379B2 (en)  20101117  20131022  Lsi Corporation  Systems and methods for self tuning target adaptation 
US8566665B2 (en)  20110624  20131022  Lsi Corporation  Systems and methods for error correction using low density parity check codes using multiple layer check equations 
US8578241B2 (en)  20111010  20131105  Lsi Corporation  Systems and methods for parity sharing data processing 
US8595576B2 (en)  20110630  20131126  Lsi Corporation  Systems and methods for evaluating and debugging LDPC iterative decoders 
US8611033B2 (en)  20110415  20131217  Lsi Corporation  Systems and methods for selective decoder input data processing 
US8630053B2 (en)  20120214  20140114  Lsi Corporation  Systems and methods for parameter modification during data processing retry 
US8661324B2 (en)  20110908  20140225  Lsi Corporation  Systems and methods for nonbinary decoding biasing control 
US8661071B2 (en)  20101011  20140225  Lsi Corporation  Systems and methods for partially conditioned noise predictive equalization 
US8667039B2 (en)  20101117  20140304  Lsi Corporation  Systems and methods for variance dependent normalization for branch metric calculation 
US8670955B2 (en)  20110415  20140311  Lsi Corporation  Systems and methods for reliability assisted noise predictive filtering 
US8683309B2 (en)  20111028  20140325  Lsi Corporation  Systems and methods for ambiguity based decode algorithm modification 
US8681441B2 (en)  20110908  20140325  Lsi Corporation  Systems and methods for generating predictable degradation bias 
US8681439B2 (en)  20100913  20140325  Lsi Corporation  Systems and methods for handling sector gaps in intertrack interference compensation 
US8689062B2 (en)  20111003  20140401  Lsi Corporation  Systems and methods for parameter selection using reliability information 
US8693120B2 (en)  20110317  20140408  Lsi Corporation  Systems and methods for sample averaging in data processing 
US8699167B2 (en)  20110216  20140415  Lsi Corporation  Systems and methods for data detection using distance based tuning 
US8707144B2 (en)  20111017  20140422  Lsi Corporation  LDPC decoder with targeted symbol flipping 
US8736998B2 (en)  20120517  20140527  Lsi Corporation  Systems and methods for symbol regrouping decoding processing 
US8750447B2 (en)  20101102  20140610  Lsi Corporation  Systems and methods for variable thresholding in a pattern detector 
US8751913B2 (en)  20111114  20140610  Lsi Corporation  Systems and methods for reduced power multilayer data decoding 
US8762815B2 (en)  20120430  20140624  Lsi Corporation  Systems and methods for data decoder state preservation during extended delay processing 
US20140181617A1 (en) *  20121220  20140626  Lsi Corporation  Management of nonvalid decision patterns of a soft read retry operation 
US8767333B2 (en)  20110922  20140701  Lsi Corporation  Systems and methods for pattern dependent target adaptation 
US8775898B2 (en)  20120517  20140708  Lsi Corporation  Systems and methods for hardware flexible low density parity check conversion 
US8775897B2 (en)  20120507  20140708  Lsi Corporation  Data processing system with failure recovery 
US8810940B2 (en)  20110207  20140819  Lsi Corporation  Systems and methods for off track error recovery 
US8819527B2 (en)  20110719  20140826  Lsi Corporation  Systems and methods for mitigating stubborn errors in a data processing system 
US8826105B2 (en)  20120412  20140902  Lsi Corporation  Data processing system with out of order transfer 
US8830613B2 (en)  20110719  20140909  Lsi Corporation  Storage media intertrack interference cancellation 
US8850276B2 (en)  20110922  20140930  Lsi Corporation  Systems and methods for efficient data shuffling in a data processing system 
US8854754B2 (en)  20110819  20141007  Lsi Corporation  Systems and methods for local iteration adjustment 
US8854753B2 (en)  20110317  20141007  Lsi Corporation  Systems and methods for auto scaling in a data processing system 
US8856631B2 (en)  20121004  20141007  Lsi Corporation  Systems and methods for parallel retry processing during iterative data processing 
US8862957B2 (en)  20120727  20141014  Lsi Corporation  Symbol selective scaling with parity forcing 
US8862960B2 (en)  20111010  20141014  Lsi Corporation  Systems and methods for parity shared data encoding 
US8862961B2 (en)  20120918  20141014  Lsi Corporation  LDPC decoder with dynamic graph modification 
US8868854B2 (en)  20111215  20141021  Lsi Corporation  Systems and methods for handling out of order reporting in a storage device 
US8879182B2 (en)  20110719  20141104  Lsi Corporation  Storage media intertrack interference cancellation 
US8887034B2 (en)  20110415  20141111  Lsi Corporation  Systems and methods for short media defect detection 
US8910005B2 (en)  20121203  20141209  Lsi Corporation  Systems and methods for selective retry data retention processing 
US8929009B2 (en)  20121219  20150106  Lsi Corporation  Irregular low density parity check decoder with low syndrome error handling 
US8930794B2 (en)  20120530  20150106  Lsi Corporation  Error injection for LDPC retry validation 
US8949696B2 (en)  20130219  20150203  Lsi Corporation  Systems and methods for trapping set disruption 
US8996969B2 (en)  20121208  20150331  Lsi Corporation  Low density parity check decoder with miscorrection handling 
US8996971B2 (en)  20120904  20150331  Lsi Corporation  LDPC decoder trapping set identification 
US9015550B2 (en)  20121005  20150421  Lsi Corporation  Low density parity check layer decoder for codes with overlapped circulants 
US9026572B2 (en)  20110829  20150505  Lsi Corporation  Systems and methods for anticausal noise predictive filtering in a data channel 
US9092368B2 (en)  20121004  20150728  Lsi Corporation  Systems and methods for modified quality based priority scheduling during iterative data processing 
US9176815B2 (en)  20131128  20151103  Seagate Technology Llc  Flash channel with selective decoder likelihood dampening 
US9190104B2 (en)  20121213  20151117  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for data retry using averaging process 
US9209835B2 (en)  20131127  20151208  Seagate Technology Llc  Read retry for nonvolatile memories 
US9219469B2 (en)  20100921  20151222  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for filter constraint estimation 
US20160026527A1 (en) *  20140723  20160128  Raidix Corporation  Systems and methods for error correction coding 
US20160043741A1 (en) *  20140806  20160211  SK Hynix Inc.  Coding method and device 
US9323625B2 (en)  20131112  20160426  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for lost synchronization data set reprocessing 
US9343082B2 (en)  20100330  20160517  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for detecting head contact 
US9385758B2 (en)  20140102  20160705  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for efficient targeted symbol flipping 
US9385756B2 (en)  20120607  20160705  Avago Technologies General Ip (Singapore) Pte. Ltd.  Data processing system with retained sector reprocessing 
US9459956B2 (en)  20130719  20161004  Seagate Technology Llc  Data decoder with trapping set flip bit mapper 
US20160373137A1 (en) *  20150616  20161222  Sk Hynix Memory Solutions Inc.  Enhanced chipkill schemes by using ecc syndrome pattern 
US9553611B2 (en) *  20141127  20170124  Apple Inc.  Error correction coding with highdegree overlap among component codes 
US20170199775A1 (en) *  20160111  20170713  Micron Technology, Inc.  Estimation of error correcting performance of lowdensity paritycheck (ldpc) codes 
Families Citing this family (2)
Publication number  Priority date  Publication date  Assignee  Title 

JP2006295510A (en) *  20050408  20061026  Mitsubishi Electric Corp  Fecframe coding equipment, efc multiplexer, fec demultiplexer, and optical communications device 
JP5699737B2 (en) *  20110328  20150415  富士通株式会社  Error correction device and an error correction method 
Citations (23)
Publication number  Priority date  Publication date  Assignee  Title 

US4336612A (en) *  19780117  19820622  Mitsubishi Denki Kabushiki Kaisha  Error correction encoding and decoding system 
US4546474A (en) *  19800620  19851008  Sony Corporation  Method of error correction 
US4607367A (en) *  19830719  19860819  Sony Corporation  Correcting errors in binary data 
US4646301A (en) *  19831031  19870224  Hitachi, Ltd.  Decoding method and system for doublyencoded ReedSolomon codes 
US4653051A (en) *  19830914  19870324  Matsushita Electric Industrial Co., Ltd.  Apparatus for detecting and correcting errors on product codes 
US4653052A (en) *  19840629  19870324  Hitachi, Ltd.  Method for decoding doubleencoding codes and apparatus using the same 
US4677622A (en) *  19830622  19870630  Hitachi, Ltd.  Error correction method and system 
US4682333A (en) *  19840628  19870721  Mitsubishi Denki Kabushiki Kaisha  Decoder for decoding a twostage encoded code 
US4683572A (en) *  19840229  19870728  U.S. Philips Corporation  Decoding device for code symbols protected by ReedSolomon code 
US4719628A (en) *  19831220  19880112  Sony Corporation  Method and apparatus for decoding error correction code 
US4782490A (en) *  19870316  19881101  Cythera Corporation  Method and a system for multiple error detection and correction 
US4785451A (en) *  19860110  19881115  Sony Corporation  Generator for an error correcting code, a decoder therefore, and a method for the same 
US4821268A (en) *  19871026  19890411  Cyclotomics, Inc.  Soft decision ReedSolomon decoder 
US4845714A (en) *  19870608  19890704  Exabyte Corporation  Multiple pass error correction process and apparatus for product codes 
US4916702A (en) *  19880617  19900410  Cyclotomics, Inc.  Elongated burst trapping 
US4972416A (en) *  19871113  19901120  Matsushita Electric Industrial Co., Ltd.  Error detection and correction method 
US5224106A (en) *  19900509  19930629  Digital Equipment Corporation  Multilevel error correction system 
US5247523A (en) *  19890712  19930921  Hitachi, Ltd.  Code error correction apparatus 
US5278846A (en) *  19900611  19940111  Matsushita Electric Industrial Co., Ltd.  Digital signal decoder 
US5381422A (en) *  19890905  19950110  Canon Kabushiki Kaisha  Device for correcting code error 
US5422895A (en) *  19920109  19950606  Quantum Corporation  Crosschecking for onthefly Reed Solomon error correction code 
US5436917A (en) *  19920518  19950725  Canon Kabushiki Kaisha  Data processing apparatus having error detection/correction codes 
US5539755A (en) *  19920519  19960723  U.S. Philips Corporation  Method and apparatus for decoding an error protected block of extended data 
Family Cites Families (1)
Publication number  Priority date  Publication date  Assignee  Title 

JP3328093B2 (en)  19940712  20020924  三菱電機株式会社  Error correction device 
Patent Citations (23)
Publication number  Priority date  Publication date  Assignee  Title 

US4336612A (en) *  19780117  19820622  Mitsubishi Denki Kabushiki Kaisha  Error correction encoding and decoding system 
US4546474A (en) *  19800620  19851008  Sony Corporation  Method of error correction 
US4677622A (en) *  19830622  19870630  Hitachi, Ltd.  Error correction method and system 
US4607367A (en) *  19830719  19860819  Sony Corporation  Correcting errors in binary data 
US4653051A (en) *  19830914  19870324  Matsushita Electric Industrial Co., Ltd.  Apparatus for detecting and correcting errors on product codes 
US4646301A (en) *  19831031  19870224  Hitachi, Ltd.  Decoding method and system for doublyencoded ReedSolomon codes 
US4719628A (en) *  19831220  19880112  Sony Corporation  Method and apparatus for decoding error correction code 
US4683572A (en) *  19840229  19870728  U.S. Philips Corporation  Decoding device for code symbols protected by ReedSolomon code 
US4682333A (en) *  19840628  19870721  Mitsubishi Denki Kabushiki Kaisha  Decoder for decoding a twostage encoded code 
US4653052A (en) *  19840629  19870324  Hitachi, Ltd.  Method for decoding doubleencoding codes and apparatus using the same 
US4785451A (en) *  19860110  19881115  Sony Corporation  Generator for an error correcting code, a decoder therefore, and a method for the same 
US4782490A (en) *  19870316  19881101  Cythera Corporation  Method and a system for multiple error detection and correction 
US4845714A (en) *  19870608  19890704  Exabyte Corporation  Multiple pass error correction process and apparatus for product codes 
US4821268A (en) *  19871026  19890411  Cyclotomics, Inc.  Soft decision ReedSolomon decoder 
US4972416A (en) *  19871113  19901120  Matsushita Electric Industrial Co., Ltd.  Error detection and correction method 
US4916702A (en) *  19880617  19900410  Cyclotomics, Inc.  Elongated burst trapping 
US5247523A (en) *  19890712  19930921  Hitachi, Ltd.  Code error correction apparatus 
US5381422A (en) *  19890905  19950110  Canon Kabushiki Kaisha  Device for correcting code error 
US5224106A (en) *  19900509  19930629  Digital Equipment Corporation  Multilevel error correction system 
US5278846A (en) *  19900611  19940111  Matsushita Electric Industrial Co., Ltd.  Digital signal decoder 
US5422895A (en) *  19920109  19950606  Quantum Corporation  Crosschecking for onthefly Reed Solomon error correction code 
US5436917A (en) *  19920518  19950725  Canon Kabushiki Kaisha  Data processing apparatus having error detection/correction codes 
US5539755A (en) *  19920519  19960723  U.S. Philips Corporation  Method and apparatus for decoding an error protected block of extended data 
Cited By (258)
Publication number  Priority date  Publication date  Assignee  Title 

JP3328093B2 (en)  19940712  20020924  三菱電機株式会社  Error correction device 
US5898531A (en) *  19940901  19990427  Cannon Kabushiki Kaisha  Signal processing apparatus for digitally modulating input data 
US5901159A (en) *  19951225  19990504  Sony Corporation  Apparatus and method for decoding a coded digital signal 
US6047398A (en) *  19960703  20000404  Matsushita Electric Industrial Co., Ltd.  Reproducing method, reproducing apparatus and recording and reproducing apparatus using the same reproducing method, and recording medium having the same method recorded therein 
US5983389A (en) *  19961008  19991109  Canon Kabushiki Kaisha  Error correction decoding apparatus 
US6108811A (en) *  19961018  20000822  Mitsubishi Denki Kabushiki Kaisha  Errorcorrecting decoder continuously adding flag signals to locations preceding a first location at which a difference between path metrics is lower than the threshold 
US6138263A (en) *  19970408  20001024  Kabushiki Kaisha Toshiba  Error correcting method and apparatus for information data having error correcting product code block 
US6131178A (en) *  19970415  20001010  Mitsubishi Denki Kabushiki Kaisha  Error correcting decoding apparatus of extended ReedSolomon code, and error correcting apparatus of singly or doubly extended ReedSolomon codes 
US6233710B1 (en) *  19970514  20010515  Texas Instruments Incorporated  ReedSolomon decoding device 
US6073250A (en) *  19971106  20000606  Luby; Michael G.  Loss resilient decoding technique 
US20050015706A1 (en) *  19971201  20050120  Macrovision Europe Limited  Codeword for use in digital optical media and a method of generating therefor 
US7181673B2 (en) *  19971201  20070220  Macrovision Europe Limited  Codeword for use in digital optical media and a method of generating therefor 
US6058500A (en) *  19980120  20000502  3Com Corporation  Highspeed syndrome calculation 
US6029186A (en) *  19980120  20000222  3Com Corporation  High speed calculation of cyclical redundancy check sums 
US6219815B1 (en)  19980120  20010417  3Com Corporation  Highspeed syndrome calculation 
US6026420A (en) *  19980120  20000215  3Com Corporation  Highspeed evaluation of polynomials 
US6412090B1 (en) *  19980618  20020625  Globespanvirata, Inc.  Galois field computation system and method 
US6625774B1 (en) *  19991029  20030923  Stmicroelectronics, Inc.  Redundancy system and method for locating errors in interleaved code words 
US6634007B1 (en)  19991108  20031014  Codevector Technology  Algebraic soft decoding of reedsolomon codes 
US6721373B1 (en) *  20000329  20040413  Tioga Technologies Ltd.  Multitone receiver and a method for operating the same 
US7752522B2 (en)  20000605  20100706  Qualcomm Incorporated  Method and apparatus for recovery of particular bits of a frame 
US7159164B1 (en) *  20000605  20070102  Qualcomm Incorporated  Method and apparatus for recovery of particular bits of a frame 
US20040194003A1 (en) *  20000609  20040930  Kwon HyungJoon  Method and apparatus for correcting C1/PI word errors using error locations detected by EFM/EFM+ decoding 
US7266748B2 (en) *  20000609  20070904  Samsung Electronics Co., Ltd.  Method and apparatus for correcting C1/PI word errors using error locations detected by EFM/EFM+ decoding 
US6721917B2 (en) *  20000616  20040413  Acer Laboratories, Inc.  Method and system for optical disk decoding 
WO2002043250A3 (en) *  20001127  20020725  Matra Nortel Communications  Method for decoding a block of symbols and device therefor 
FR2817418A1 (en) *  20001127  20020531  Matra Nortel Communications  A method of decoding a block of symbols and a device implementing such a process 
WO2002043250A2 (en) *  20001127  20020530  Nortel Networks France  Method for decoding a block of symbols and device therefor 
US20040030984A1 (en) *  20001127  20040212  Christine Renaud  Method for decoding a block of symbols and device therefor 
US20040078745A1 (en) *  20010218  20040422  Martin Kruger  Method and device for error correction of data blocks 
US7028243B2 (en) *  20010228  20060411  Infineon Technologies Ag  Method and device for error correction of data blocks 
EP1468498A4 (en) *  20020123  20060531  Thomson Licensing  Chien search cell for an errorcorrecting decoder 
EP1468499A4 (en) *  20020123  20051214  Thomson Licensing  Intradecoder component block messaging 
EP1468499A1 (en) *  20020123  20041020  Thomson Licensing S.A.  Intradecoder component block messaging 
EP1470647A1 (en) *  20020123  20041027  Thomson Licensing S.A.  Dual chien search blocks in an errorcorrecting decoder 
US20030140301A1 (en) *  20020123  20030724  Litwin Louis Robert  Intradecoder component block messaging 
US7020826B2 (en)  20020123  20060328  Thomson Licensing  Intradecoder component block messaging 
EP1470647A4 (en) *  20020123  20051214  Thomson Licensing  Dual chien search blocks in an errorcorrecting decoder 
EP1468498A1 (en) *  20020123  20041020  Thomson Licensing S.A.  Chien search cell for an errorcorrecting decoder 
US7171594B2 (en) *  20020306  20070130  HewlettPackard Development Company, L.P.  Pausing a transfer of data 
US20030172328A1 (en) *  20020306  20030911  Wyatt Stewart R.  Pausing a transfer of data 
US20050229085A1 (en) *  20020405  20051013  Koninklijke Philips Electronics N.V.  Method and apparatus for embedding an additional layer of error correction into an error correcting code 
US7340663B2 (en) *  20020405  20080304  Koninklijke Philiops Electronics N.V.  Method and apparatus for embedding an additional layer of error correction into an error correcting code 
US20040003334A1 (en) *  20020630  20040101  Boyanapalli Vipin S.  Efficient method and apparatus for low latency forward error correction 
US7007223B2 (en) *  20020630  20060228  Intel Corporation  Efficient method and apparatus for low latency forward error correction 
US7131052B2 (en) *  20020812  20061031  International Business Machines Corporation  Algebraic decoder and method for correcting an arbitrary mixture of burst and random errors 
US20040030737A1 (en) *  20020812  20040212  International Business Machines Corporation  Algebraic decoder and method for correcting an arbitrary mixture of burst and random errors 
US20040059985A1 (en) *  20020925  20040325  Sharp Guy Bl  Method and apparatus for tracking address of memory errors 
US7243292B1 (en) *  20021017  20070710  Telefonaktiebolaget Lm Ericsson (Publ)  Error correction using finite fields of odd characteristics on binary hardware 
US7197527B2 (en) *  20021017  20070327  Telefonaktiebolaget Lm Ericsson (Publ)  Efficient arithmetic in finite fields of odd characteristic on binary hardware 
US20070150794A1 (en) *  20021017  20070628  Mats Naslund  Error correction using finite fields of odd characteristic on binary hardware 
US20040078407A1 (en) *  20021017  20040422  Mats Naslund  Efficient arithmetic in finite fields of odd characteristic on binary hardware 
US20040193966A1 (en) *  20030313  20040930  Osamu Hirabayashi  Semiconductor device 
US7823046B2 (en) *  20030313  20101026  Kabushiki Kaisha Toshiba  Semiconductor device 
US7181669B2 (en) *  20030603  20070220  Sunplus Technology Co., Ltd.  Device and method for block code error correction 
US20040249840A1 (en) *  20030603  20041209  Sunplus Technology Co., Ltd.  Device and method for block code error correction 
US7954040B2 (en) *  20031128  20110531  Mediatek Inc.  Encoding method and apparatus for cross interleaved cyclic codes 
US20080022192A1 (en) *  20031128  20080124  Mediatek Inc.  Encoding method and apparatus for cross interleaved cyclic codes 
US7472333B2 (en) *  20031128  20081230  Mediatek, Inc.  Encoding method and apparatus for cross interleaved cyclic codes 
US20050120285A1 (en) *  20031128  20050602  Mediatek Inc.  Encoding method and apparatus for cross interleaved cyclic codes 
WO2005101207A1 (en) *  20040414  20051027  Koninklijke Philips Electronics N.V.  Data handling device that corrects errors in a data memory 
US20070277083A1 (en) *  20040414  20071129  Koninklijke Philips Electronics, N.V.  Data Handling Device That Corects Errors In A Data Memory 
US8689077B2 (en) *  20040506  20140401  Micron Technology, Inc.  Memory controller method and system compensating for memory cell data losses 
US9064600B2 (en) *  20040506  20150623  Micron Technology, Inc.  Memory controller method and system compensating for memory cell data losses 
US7447974B2 (en) *  20040506  20081104  Micron Technology, Inc.  Memory controller method and system compensating for memory cell data losses 
US20140181613A1 (en) *  20040506  20140626  Micron Technology, Inc.  Memory controller method and system compensating for memory cell data losses 
US7447973B2 (en) *  20040506  20081104  Micron Technology, Inc.  Memory controller method and system compensating for memory cell data losses 
US7428687B2 (en) *  20040506  20080923  Micron Technology, Inc.  Memory controller method and system compensating for memory cell data losses 
US20090024884A1 (en) *  20040506  20090122  Micron Technology, Inc.  Memory controller method and system compensating for memory cell data losses 
US20060069856A1 (en) *  20040506  20060330  Klein Dean A  Memory controller method and system compensating for memory cell data losses 
US20060056259A1 (en) *  20040506  20060316  Klein Dean A  Memory controller method and system compensating for memory cell data losses 
US7836374B2 (en)  20040506  20101116  Micron Technology, Inc.  Memory controller method and system compensating for memory cell data losses 
US20110060961A1 (en) *  20040506  20110310  Micro Technology, Inc.  Memory controller method and system compensating for memory cell data losses 
US20060056260A1 (en) *  20040506  20060316  Klein Dean A  Memory controller method and system compensating for memory cell data losses 
US7461320B2 (en)  20040624  20081202  Micron Technology, Inc.  Memory system and method having selective ECC during low power refresh 
US20060206769A1 (en) *  20040624  20060914  Klein Dean A  Memory system and method having selective ECC during low power refresh 
US7340668B2 (en)  20040625  20080304  Micron Technology, Inc.  Low power costeffective ECC memory system and method 
US7526713B2 (en)  20040625  20090428  Micron Technology, Inc.  Low power costeffective ECC memory system and method 
US20050289444A1 (en) *  20040625  20051229  Klein Dean A  Low power costeffective ECC memory system and method 
US20060218469A1 (en) *  20040625  20060928  Klein Dean A  Low power costeffective ECC memory system and method 
US8279683B2 (en)  20040715  20121002  Micron Technology, Inc.  Digit line comparison circuits 
US7558142B2 (en)  20040715  20090707  Micron Technology, Inc.  Method and system for controlling refresh to avoid memory cell data losses 
US20100054070A1 (en) *  20040715  20100304  Micron Technology, Inc.  Method and system for controlling refresh to avoid memory cell data losses 
US20110134708A1 (en) *  20040715  20110609  Micron Technology, Inc.  Method and system for controlling refresh to avoid memory cell data losses 
US8446783B2 (en)  20040715  20130521  Micron Technology, Inc.  Digit line comparison circuits 
US20080002503A1 (en) *  20040715  20080103  Klein Dean A  Method and system for controlling refresh to avoid memory cell data losses 
US7898892B2 (en)  20040715  20110301  Micron Technology, Inc.  Method and system for controlling refresh to avoid memory cell data losses 
US7623392B2 (en)  20040715  20091124  Micron Technology, Inc.  Method and system for controlling refresh to avoid memory cell data losses 
US8601343B2 (en)  20040727  20131203  Nokia Siemens Networks Gmbh & Co. Kg  Encoding and decoding method, and encoding and decoding devices with a twostage error protection process 
US20080320358A1 (en) *  20040727  20081225  Jurgen Pandel  Encoding and Decoding Method, and Encoding and Decoding Devices with a TwoStage Error Protection Process 
US20110131474A1 (en) *  20040727  20110602  Nokia Siemens Networks Gmbh & Co. Kg  Encoding and decoding method, and encoding and decoding devices with a twostage error protection process 
US7647546B2 (en)  20041222  20100112  Samsung Electronics Co., Ltd.  Apparatus and method for decoding ReedSolomon symbols 
KR100871206B1 (en) *  20041222  20081201  삼성전자주식회사  Apparatus and method for decoding of reedsolomon symbols 
US20060150066A1 (en) *  20041222  20060706  Samsung Electronics Co., Ltd.  Apparatus and method for decoding Reedsolomon symbols 
WO2006068436A1 (en) *  20041222  20060629  Samsung Electronics Co., Ltd.  Apparatus and method for decoding reedsolomon symbols 
US20060150054A1 (en) *  20050103  20060706  JiaPing Chen  Decoding device for decoding product code and decoding method using the same 
US7395491B2 (en) *  20050103  20080701  Sunplus Technology Co., Ltd.  Decoding device for decoding product code and decoding method using the same 
US20060174182A1 (en) *  20050128  20060803  Neefs Henk G C  Trialanderror multibit error correction 
US7607071B2 (en) *  20050128  20091020  Intel Corporation  Error correction using iterating generation of data syndrome 
US7836385B2 (en) *  20050203  20101116  Broadcom Corporation  MetaViterbi algorithm for use in communication systems 
US20090228769A1 (en) *  20050203  20090910  Vityaev Andrei E  Metaviterbi algorithm for use in communication systems 
US7370258B2 (en) *  20050428  20080506  Sandbridge Technologies Inc.  Iterative concatenated convolutional ReedSolomon decoding method 
US20060248430A1 (en) *  20050428  20061102  Sandbridge Technologies Inc.  Iterative concatenated convolutional ReedSolomon decoding method 
US20090292975A1 (en) *  20050429  20091126  Agere Systems Inc.  Method and apparatus for iterative errorerasure decoding 
US8930797B2 (en) *  20050429  20150106  Lsi Corporation  Method and apparatus for iterative errorerasure decoding 
US8832523B2 (en) *  20060303  20140909  Ternarylogic Llc  Multistate symbol error correction in matrix based codes 
US20090172501A1 (en) *  20060303  20090702  Ternarylogic Llc  MultiState Symbol Error Correction in Matrix Based Codes 
US8250314B2 (en) *  20060329  20120821  Micron Technology, Inc.  Method and device for checking sector protection information of a nonvolatile memory 
US20120311415A1 (en) *  20060329  20121206  Micron Technology, Inc.  Method and device for detecting possible corruption of sector protection information of a nonvolatile memory stored in an on board volatile memory array at poweron 
US20070233984A1 (en) *  20060329  20071004  Stmicroelectronics S.R.L.  Method and device for detecting possible corruption of sector protection information of a nonvolatile memory stored in an on board volatile memory array at poweron 
US9203436B2 (en) *  20060712  20151201  Ternarylogic Llc  Error correction in multivalued (p,k) codes 
US20080016432A1 (en) *  20060712  20080117  Peter Lablans  Error Correction in MultiValued (p,k) Codes 
US20090249164A1 (en) *  20060802  20091001  Jorg Hammer  Method for serial asynchronous transmission of data in an arragement for the monitoring, controlling, and regulating an operational control facility of building 
US8037379B1 (en) *  20060831  20111011  Pdf Solutions, Inc.  Prediction of impact on postrepair yield resulting from manufacturing process modification 
WO2008045292A2 (en) *  20061004  20080417  Marvell World Trade Ltd.  Error correction decoding methods and apparatus 
WO2008045292A3 (en) *  20061004  20080703  Marvell World Trade Ltd  Error correction decoding methods and apparatus 
US8032812B1 (en)  20061004  20111004  Marvell World Trade Ltd.  Error correction decoding methods and apparatus 
US20080092016A1 (en) *  20061011  20080417  Micron Technology, Inc.  Memory system and method using partial ECC to achieve low power refresh and fast access to data 
US9286161B2 (en)  20061011  20160315  Micron Technology, Inc.  Memory system and method using partial ECC to achieve low power refresh and fast access to data 
US8832522B2 (en)  20061011  20140909  Micron Technology, Inc.  Memory system and method using partial ECC to achieve low power refresh and fast access to data 
US7894289B2 (en)  20061011  20110222  Micron Technology, Inc.  Memory system and method using partial ECC to achieve low power refresh and fast access to data 
US20110138251A1 (en) *  20061011  20110609  Pawlowski J Thomas  Memory system and method using partial ecc to achieve low power refresh and fast access to data 
US8359517B2 (en)  20061011  20130122  Micron Technology, Inc.  Memory system and method using partial ECC to achieve low power refresh and fast access to data 
US8880974B2 (en)  20061018  20141104  Micron Technology, Inc.  Memory system and method using ECC with flag bit to identify modified data 
US20110138252A1 (en) *  20061018  20110609  Pawlowski J Thomas  Memory system and method using ecc with flag bit to identify modified data 
US8413007B2 (en)  20061018  20130402  Micron Technology, Inc.  Memory system and method using ECC with flag bit to identify modified data 
US8601341B2 (en)  20061018  20131203  Micron Technologies, Inc.  Memory system and method using ECC with flag bit to identify modified data 
US7900120B2 (en)  20061018  20110301  Micron Technology, Inc.  Memory system and method using ECC with flag bit to identify modified data 
US20080109705A1 (en) *  20061018  20080508  Pawlowski J Thomas  Memory system and method using ECC with flag bit to identify modified data 
US8020072B2 (en) *  20061025  20110913  International Business Machines Corporation  Method and apparatus for correcting data errors 
US20080109687A1 (en) *  20061025  20080508  Christopher Michael Abernathy  Method and apparatus for correcting data errors 
US7949927B2 (en) *  20061114  20110524  Samsung Electronics Co., Ltd.  Error correction method and apparatus for predetermined error patterns 
US20080115041A1 (en) *  20061114  20080515  Jihoon Park  Error correction method and apparatus for predetermined error patterns 
US20080168330A1 (en) *  20070108  20080710  Agere Systems Inc.  Systems and methods for prioritizing error correction data 
US7971125B2 (en) *  20070108  20110628  Agere Systems Inc.  Systems and methods for prioritizing error correction data 
US10141950B2 (en)  20070501  20181127  The Texas A&M University System  Low density parity check decoder 
US8555140B2 (en)  20070501  20131008  The Texas A&M University System  Low density parity check decoder for irregular LDPC codes 
US8656250B2 (en)  20070501  20140218  Texas A&M University System  Low density parity check decoder for regular LDPC codes 
US9112530B2 (en)  20070501  20150818  The Texas A&M University System  Low density parity check decoder 
US8359522B2 (en)  20070501  20130122  Texas A&M University System  Low density parity check decoder for regular LDPC codes 
US8418023B2 (en)  20070501  20130409  The Texas A&M University System  Low density parity check decoder for irregular LDPC codes 
US7900117B2 (en)  20070926  20110301  Kabushiki Kaisha Toshiba  Semiconductor memory device and its control method 
US8127212B2 (en) *  20070926  20120228  HewlettPackard Development Company, L.P.  System and method for determining the faulttolerance of an erasure code 
US20090177944A1 (en) *  20070926  20090709  Shinichi Kanno  Semiconductor memory device and its control method 
US20090083590A1 (en) *  20070926  20090326  John Johnson Wylie  System and method for determining the faulttolerance of an erasure code 
KR100996770B1 (en)  20070926  20101125  가부시끼가이샤 도시바  Semiconductor memory device and its control method 
WO2009041153A1 (en) *  20070926  20090402  Kabushiki Kaisha Toshiba  Semiconductor memory device and its control method 
US20110047441A1 (en) *  20080301  20110224  Kabushiki Kaisha Toshiba  Chien search device and chien search method 
US8418028B2 (en) *  20080301  20130409  Kabushiki Kaisha Toshiba  Chien search device and Chien search method 
US8468418B2 (en)  20080502  20130618  Lsi Corporation  Systems and methods for queue based data detection and decoding 
US20090282313A1 (en) *  20080507  20091112  Legend Silicon Corp.  Method and apparatus for hard decision bounded distance decoding of nordstrom robinson code 
US8250427B2 (en)  20080618  20120821  HewlettPackard Development Company  Selecting erasure codes for a fault tolerant system 
US20100083069A1 (en) *  20080618  20100401  John Johnson Wylie  Selecting Erasure Codes For A Fault Tolerant System 
US20110080211A1 (en) *  20081120  20110407  Shaohua Yang  Systems and Methods for Noise Reduced Data Detection 
US8413020B2 (en)  20090812  20130402  Lsi Corporation  Systems and methods for retimed virtual data processing 
US20120260150A1 (en) *  20091217  20121011  International Business Machines Corporation  Data management in solid state storage systems 
CN102656566A (en) *  20091217  20120905  国际商业机器公司  Data management in solid state storage systems 
US9037951B2 (en) *  20091217  20150519  International Business Machines Corporation  Data management in solid state storage systems 
US8743936B2 (en)  20100105  20140603  Lsi Corporation  Systems and methods for determining noise components in a signal set 
US20110164669A1 (en) *  20100105  20110707  Lsi Corporation  Systems and Methods for Determining Noise Components in a Signal Set 
US9343082B2 (en)  20100330  20160517  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for detecting head contact 
US8161351B2 (en)  20100330  20120417  Lsi Corporation  Systems and methods for efficient data storage 
US8418019B2 (en)  20100419  20130409  Lsi Corporation  Systems and methods for dynamic scaling in a data decoding system 
US8443249B2 (en)  20100426  20130514  Lsi Corporation  Systems and methods for low density parity check data encoding 
US8527831B2 (en)  20100426  20130903  Lsi Corporation  Systems and methods for low density parity check data decoding 
US8381071B1 (en)  20100521  20130219  Lsi Corporation  Systems and methods for decoder sharing between data sets 
US8381074B1 (en)  20100521  20130219  Lsi Corporation  Systems and methods for utilizing a centralized queue based data processing circuit 
US8208213B2 (en)  20100602  20120626  Lsi Corporation  Systems and methods for hybrid algorithm gain adaptation 
US8773794B2 (en)  20100913  20140708  Lsi Corporation  Systems and methods for blockwise intertrack interference compensation 
US8804260B2 (en)  20100913  20140812  Lsi Corporation  Systems and methods for intertrack interference compensation 
US8681439B2 (en)  20100913  20140325  Lsi Corporation  Systems and methods for handling sector gaps in intertrack interference compensation 
US8295001B2 (en)  20100921  20121023  Lsi Corporation  Systems and methods for low latency noise cancellation 
US9219469B2 (en)  20100921  20151222  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for filter constraint estimation 
US8385014B2 (en)  20101011  20130226  Lsi Corporation  Systems and methods for identifying potential media failure 
US8560930B2 (en)  20101011  20131015  Lsi Corporation  Systems and methods for multilevel quasicyclic low density parity check codes 
US8661071B2 (en)  20101011  20140225  Lsi Corporation  Systems and methods for partially conditioned noise predictive equalization 
US8443250B2 (en)  20101011  20130514  Lsi Corporation  Systems and methods for error correction using irregular low density parity check codes 
US8750447B2 (en)  20101102  20140610  Lsi Corporation  Systems and methods for variable thresholding in a pattern detector 
US8667039B2 (en)  20101117  20140304  Lsi Corporation  Systems and methods for variance dependent normalization for branch metric calculation 
US8566379B2 (en)  20101117  20131022  Lsi Corporation  Systems and methods for self tuning target adaptation 
US8810940B2 (en)  20110207  20140819  Lsi Corporation  Systems and methods for off track error recovery 
US8699167B2 (en)  20110216  20140415  Lsi Corporation  Systems and methods for data detection using distance based tuning 
US8446683B2 (en)  20110222  20130521  Lsi Corporation  Systems and methods for data precoding calibration 
US8854753B2 (en)  20110317  20141007  Lsi Corporation  Systems and methods for auto scaling in a data processing system 
US8693120B2 (en)  20110317  20140408  Lsi Corporation  Systems and methods for sample averaging in data processing 
US8611033B2 (en)  20110415  20131217  Lsi Corporation  Systems and methods for selective decoder input data processing 
US8887034B2 (en)  20110415  20141111  Lsi Corporation  Systems and methods for short media defect detection 
US8670955B2 (en)  20110415  20140311  Lsi Corporation  Systems and methods for reliability assisted noise predictive filtering 
US8499231B2 (en)  20110624  20130730  Lsi Corporation  Systems and methods for reduced format nonbinary decoding 
US8566665B2 (en)  20110624  20131022  Lsi Corporation  Systems and methods for error correction using low density parity check codes using multiple layer check equations 
US8560929B2 (en)  20110624  20131015  Lsi Corporation  Systems and methods for nonbinary decoding 
US8595576B2 (en)  20110630  20131126  Lsi Corporation  Systems and methods for evaluating and debugging LDPC iterative decoders 
US8566666B2 (en)  20110711  20131022  Lsi Corporation  Minsum based nonbinary LDPC decoder 
US8879182B2 (en)  20110719  20141104  Lsi Corporation  Storage media intertrack interference cancellation 
US8830613B2 (en)  20110719  20140909  Lsi Corporation  Storage media intertrack interference cancellation 
US8819527B2 (en)  20110719  20140826  Lsi Corporation  Systems and methods for mitigating stubborn errors in a data processing system 
US8539328B2 (en)  20110819  20130917  Lsi Corporation  Systems and methods for noise injection driven parameter selection 
US8854754B2 (en)  20110819  20141007  Lsi Corporation  Systems and methods for local iteration adjustment 
US9026572B2 (en)  20110829  20150505  Lsi Corporation  Systems and methods for anticausal noise predictive filtering in a data channel 
US8661324B2 (en)  20110908  20140225  Lsi Corporation  Systems and methods for nonbinary decoding biasing control 
US8681441B2 (en)  20110908  20140325  Lsi Corporation  Systems and methods for generating predictable degradation bias 
US8850276B2 (en)  20110922  20140930  Lsi Corporation  Systems and methods for efficient data shuffling in a data processing system 
US8767333B2 (en)  20110922  20140701  Lsi Corporation  Systems and methods for pattern dependent target adaptation 
US8689062B2 (en)  20111003  20140401  Lsi Corporation  Systems and methods for parameter selection using reliability information 
US8479086B2 (en)  20111003  20130702  Lsi Corporation  Systems and methods for efficient parameter modification 
US8578241B2 (en)  20111010  20131105  Lsi Corporation  Systems and methods for parity sharing data processing 
US8862960B2 (en)  20111010  20141014  Lsi Corporation  Systems and methods for parity shared data encoding 
US8707144B2 (en)  20111017  20140422  Lsi Corporation  LDPC decoder with targeted symbol flipping 
US8527858B2 (en)  20111028  20130903  Lsi Corporation  Systems and methods for selective decode algorithm modification 
US8683309B2 (en)  20111028  20140325  Lsi Corporation  Systems and methods for ambiguity based decode algorithm modification 
US8443271B1 (en)  20111028  20130514  Lsi Corporation  Systems and methods for dual process data decoding 
US20130110962A1 (en) *  20111101  20130502  Cleversafe, Inc.  Wirelessly Sending a Set of Encoded Data Slices 
US9798616B2 (en) *  20111101  20171024  International Business Machines Corporation  Wireless sending a set of encoded data slices 
US8751913B2 (en)  20111114  20140610  Lsi Corporation  Systems and methods for reduced power multilayer data decoding 
US8531320B2 (en)  20111114  20130910  Lsi Corporation  Systems and methods for memory efficient data decoding 
US8443251B1 (en)  20111215  20130514  Lsi Corporation  Systems and methods for out of order processing in a data retry 
US8868854B2 (en)  20111215  20141021  Lsi Corporation  Systems and methods for handling out of order reporting in a storage device 
US8762807B2 (en)  20111215  20140624  Lsi Corporation  Systems and methods for out of order processing in a data retry 
US8630053B2 (en)  20120214  20140114  Lsi Corporation  Systems and methods for parameter modification during data processing retry 
US9448884B2 (en) *  20120305  20160920  Micron Technology, Inc.  Apparatuses and methods for encoding using error protection codes 
US20160364293A1 (en) *  20120305  20161215  Micron Technology, Inc.  Apparatuses and methods for encoding using error protection codes 
US10133628B2 (en) *  20120305  20181120  Micron Technology, Inc.  Apparatuses and methods for encoding using error protection codes 
US8788915B2 (en) *  20120305  20140722  Micron Technology, Inc.  Apparatuses and methods for encoding using error protection codes 
US20130232396A1 (en) *  20120305  20130905  Micron Technology Inc.  Apparatuses and methods for encoding using error protection codes 
US20140365846A1 (en) *  20120305  20141211  Micron Technology, Inc.  Apparatuses and Methods for Encoding Using Error Protection Codes 
US8826105B2 (en)  20120412  20140902  Lsi Corporation  Data processing system with out of order transfer 
US8762815B2 (en)  20120430  20140624  Lsi Corporation  Systems and methods for data decoder state preservation during extended delay processing 
US8775897B2 (en)  20120507  20140708  Lsi Corporation  Data processing system with failure recovery 
US8525707B1 (en)  20120517  20130903  Lsi Corporation  Systems and methods for dual binary and nonbinary decoding processing 
US8736998B2 (en)  20120517  20140527  Lsi Corporation  Systems and methods for symbol regrouping decoding processing 
US8775898B2 (en)  20120517  20140708  Lsi Corporation  Systems and methods for hardware flexible low density parity check conversion 
US8930794B2 (en)  20120530  20150106  Lsi Corporation  Error injection for LDPC retry validation 
US9385756B2 (en)  20120607  20160705  Avago Technologies General Ip (Singapore) Pte. Ltd.  Data processing system with retained sector reprocessing 
US8862957B2 (en)  20120727  20141014  Lsi Corporation  Symbol selective scaling with parity forcing 
US8996971B2 (en)  20120904  20150331  Lsi Corporation  LDPC decoder trapping set identification 
US8862961B2 (en)  20120918  20141014  Lsi Corporation  LDPC decoder with dynamic graph modification 
US9092368B2 (en)  20121004  20150728  Lsi Corporation  Systems and methods for modified quality based priority scheduling during iterative data processing 
US8856631B2 (en)  20121004  20141007  Lsi Corporation  Systems and methods for parallel retry processing during iterative data processing 
US9015550B2 (en)  20121005  20150421  Lsi Corporation  Low density parity check layer decoder for codes with overlapped circulants 
US8910005B2 (en)  20121203  20141209  Lsi Corporation  Systems and methods for selective retry data retention processing 
US8996969B2 (en)  20121208  20150331  Lsi Corporation  Low density parity check decoder with miscorrection handling 
US9190104B2 (en)  20121213  20151117  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for data retry using averaging process 
US8929009B2 (en)  20121219  20150106  Lsi Corporation  Irregular low density parity check decoder with low syndrome error handling 
US9015554B2 (en) *  20121220  20150421  Seagate Technology Llc  Management of nonvalid decision patterns of a soft read retry operation 
US20140181617A1 (en) *  20121220  20140626  Lsi Corporation  Management of nonvalid decision patterns of a soft read retry operation 
US8949696B2 (en)  20130219  20150203  Lsi Corporation  Systems and methods for trapping set disruption 
US9459956B2 (en)  20130719  20161004  Seagate Technology Llc  Data decoder with trapping set flip bit mapper 
US9323625B2 (en)  20131112  20160426  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for lost synchronization data set reprocessing 
US9209835B2 (en)  20131127  20151208  Seagate Technology Llc  Read retry for nonvolatile memories 
US9176815B2 (en)  20131128  20151103  Seagate Technology Llc  Flash channel with selective decoder likelihood dampening 
US9385758B2 (en)  20140102  20160705  Avago Technologies General Ip (Singapore) Pte. Ltd.  Systems and methods for efficient targeted symbol flipping 
US20160026527A1 (en) *  20140723  20160128  Raidix Corporation  Systems and methods for error correction coding 
US9647690B2 (en) *  20140723  20170509  Raidix Corporation  Systems and methods for error correction coding 
US20160043741A1 (en) *  20140806  20160211  SK Hynix Inc.  Coding method and device 
US9553611B2 (en) *  20141127  20170124  Apple Inc.  Error correction coding with highdegree overlap among component codes 
US20160373137A1 (en) *  20150616  20161222  Sk Hynix Memory Solutions Inc.  Enhanced chipkill schemes by using ecc syndrome pattern 
US9979417B2 (en) *  20150616  20180522  SK Hynix Inc.  Enhanced chipkill schemes by using ECC syndrome pattern 
US20170199775A1 (en) *  20160111  20170713  Micron Technology, Inc.  Estimation of error correcting performance of lowdensity paritycheck (ldpc) codes 
Also Published As
Publication number  Publication date  Type 

JPH08149018A (en)  19960607  application 
JP3328093B2 (en)  20020924  grant 
Similar Documents
Publication  Publication Date  Title 

US3668631A (en)  Error detection and correction system with statistically optimized data recovery  
US3668632A (en)  Fast decode character error detection and correction system  
US4777635A (en)  ReedSolomon code encoder and syndrome generator circuit  
US5136592A (en)  Error detection and correction system for long burst errors  
US5592404A (en)  Versatile error correction system  
US6003151A (en)  Error correction and detection system for mass storage controller  
US5999959A (en)  Galois field multiplier  
US4354269A (en)  Apparatus for the processing of an information stream with the aid of an errorcorrecting convolutional code and apparatus for the detection of an error still irremediable in this processing  
US5247523A (en)  Code error correction apparatus  
US4862463A (en)  Error correcting code for 8bitperchip memory with reduced redundancy  
US5942005A (en)  Method and means for computationally efficient error and erasure correction in linear cyclic codes  
US5727003A (en)  Method and apparatus for flash burst error correction  
US5107503A (en)  High bandwidth reedsolomon encoding, decoding and error correcting circuit  
US6637002B1 (en)  Decoder for error correcting block codes  
US4821268A (en)  Soft decision ReedSolomon decoder  
US4494234A (en)  Onthefly multibyte error correcting system  
US4653052A (en)  Method for decoding doubleencoding codes and apparatus using the same  
US5602857A (en)  Error correction method and apparatus  
US4683572A (en)  Decoding device for code symbols protected by ReedSolomon code  
US4928280A (en)  Fast processor for multibit error correction codes  
US7278085B1 (en)  Simple errorcorrection codes for data buffers  
US5631914A (en)  Error correcting apparatus  
US5642367A (en)  Finite field polynomial processing module for error control coding  
US6119262A (en)  Method and apparatus for solving key equation polynomials in decoding error correction codes  
US20060123318A1 (en)  Method and apparatus for decoding low density parity check code using united node processing 
Legal Events
Date  Code  Title  Description 

AS  Assignment 
Owner name: MITSUBISHI DENKI KABUSHIKI KAISHA, JAPAN Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:INOUE, TOHRU;ONISHI, KEN;INOUE, SADAYUKI;AND OTHERS;REEL/FRAME:007583/0448;SIGNING DATES FROM 19950531 TO 19950615 

FPAY  Fee payment 
Year of fee payment: 4 

REMI  Maintenance fee reminder mailed  
LAPS  Lapse for failure to pay maintenance fees  
FP  Expired due to failure to pay maintenance fee 
Effective date: 20060127 